WO2001077871A1 - Enhanced temporal and resolution layering in advanced television - Google Patents

Enhanced temporal and resolution layering in advanced television Download PDF

Info

Publication number
WO2001077871A1
WO2001077871A1 PCT/US2001/011204 US0111204W WO0177871A1 WO 2001077871 A1 WO2001077871 A1 WO 2001077871A1 US 0111204 W US0111204 W US 0111204W WO 0177871 A1 WO0177871 A1 WO 0177871A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
ofthe
frame
resolution
digital video
Prior art date
Application number
PCT/US2001/011204
Other languages
French (fr)
Inventor
Gary E. Demos
Original Assignee
Demografx
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US09/545,233 external-priority patent/US6728317B1/en
Application filed by Demografx filed Critical Demografx
Priority to JP2001574651A priority Critical patent/JP4352105B2/en
Priority to AU2001251386A priority patent/AU2001251386A1/en
Priority to EP01924762A priority patent/EP1279111A4/en
Priority to CA002406459A priority patent/CA2406459C/en
Publication of WO2001077871A1 publication Critical patent/WO2001077871A1/en

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/14—Picture signal circuitry for video frequency region
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/13—Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC]
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
    • H04N19/17—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/30—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using hierarchical techniques, e.g. scalability
    • H04N19/39—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using hierarchical techniques, e.g. scalability involving multiple description coding [MDC], i.e. with separate layers being structured as independently decodable descriptions of input picture data
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51—Motion estimation or motion compensation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51—Motion estimation or motion compensation
    • H04N19/523—Motion estimation or motion compensation with sub-pixel accuracy
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51—Motion estimation or motion compensation
    • H04N19/55—Motion estimation with spatial constraints, e.g. at image or region borders
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51—Motion estimation or motion compensation
    • H04N19/56—Motion estimation with initialisation of the vector search, e.g. estimating a good candidate to initiate a search
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51—Motion estimation or motion compensation
    • H04N19/567—Motion estimation based on rate distortion criteria
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/587—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal sub-sampling or interpolation, e.g. decimation or subsequent interpolation of pictures in a video sequence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/59—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving spatial sub-sampling or interpolation, e.g. alteration of picture size or resolution
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
    • H04N19/61—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding in combination with predictive coding
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/80—Details of filtering operations specially adapted for video compression, e.g. for pixel interpolation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/14—Picture signal circuitry for video frequency region
    • H04N5/21—Circuitry for suppressing or minimising disturbance, e.g. moiré or halo
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/01—Conversion of standards, e.g. involving analogue television standards or digital television standards processed at pixel level
    • H04N7/0112—Conversion of standards, e.g. involving analogue television standards or digital television standards processed at pixel level one of the standards corresponding to a cinematograph film standard
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/01—Conversion of standards, e.g. involving analogue television standards or digital television standards processed at pixel level
    • H04N7/0117—Conversion of standards, e.g. involving analogue television standards or digital television standards processed at pixel level involving conversion of the spatial resolution of the incoming video signal
    • H04N7/012—Conversion between an interlaced and a progressive signal
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/01—Conversion of standards, e.g. involving analogue television standards or digital television standards processed at pixel level
    • H04N7/0127—Conversion of standards, e.g. involving analogue television standards or digital television standards processed at pixel level by changing the field or frame frequency of the incoming video signal, e.g. frame rate converter
    • H04N7/0132—Conversion of standards, e.g. involving analogue television standards or digital television standards processed at pixel level by changing the field or frame frequency of the incoming video signal, e.g. frame rate converter the field or frame frequency of the incoming video signal being multiplied by a positive integer, e.g. for flicker reduction
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N9/00—Details of colour television systems
    • H04N9/64—Circuits for processing colour signals
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/14—Picture signal circuitry for video frequency region
    • H04N5/144—Movement detection
    • H04N5/145—Movement estimation

Definitions

  • This invention relates to electronic communication systems, and more particularly to an advanced electronic television system having temporal and resolution layering of compressed image frames having enhanced compression, filtering, and display characteristics.
  • interlace is required, due to a claimed need to have about 1000 lines of resolution at high frame rates, but based upon the notion that such images cannot be compressed within the available 18-19 o mbits/second of a conventional 6 MHz broadcast television channel.
  • the present invention provides such enhancements.
  • the invention provides a method and apparatus for image compression which demonstrably achieves better than 1000-line resolution image compression at high frame rates with high quality. It also achieves both temporal and resolution scalability at this resolution at high frame rates within the available bandwidth of a conventional television broadcast channel.
  • the inventive technique efficiently achieves over twice the compression ratio being proposed for advanced television.
  • layered compression allows a form of modularized decomposition of an image that supports flexible application of a variety of image enhancement techniques.
  • Image material is preferably captured at an initial or primary framing rate of 72 fps.
  • An MPEG-like (e.g., MPEG-2, MPEG-4, etc.) data stream is then generated, comprising:
  • a base layer preferably encoded using only MPEG-type P frames, comprising a low resolution (e.g., 1024x512 pixels), low frame rate (24 or 36 Hz) bitstream;
  • an optional base resolution temporal enhancement layer encoded using only MPEG-type B frames, comprising a low resolution (e.g., 1024x512 pixels), high frame rate (72 Hz) bitstream;
  • an optional base temporal high resolution enhancement layer preferably encoded using only MPEG-type P frames, comprising a high resolution (e.g., 2kxlk pixels), low frame rate (24 or 36 Hz) bitstream;
  • an optional high resolution temporal enhancement layer encoded using only MPEG-type B frames, comprising a high resolution (e.g., 2kxlk pixels), high frame rate (72 Hz) bitstream.
  • a high resolution e.g., 2kxlk pixels
  • high frame rate 72 Hz
  • the invention provides a number of key technical attributes, allowing substantial improvement over current proposals, and including: replacement of numerous resolutions and frame rates with a single layered resolution and frame rate; no need for interlace in order to achieve better than 1000-lines of resolution for 2 megapixel images at high frame rates (72 Hz) within a 6 MHz television channel; compatibility with computer displays through use of a primary framing rate of 72 fps; and greater robustness than the current unlayered format proposal for advanced television, since all available bits may be allocated to a lower resolution base layer when "stressful" image material is encountered.
  • the invention provides a number of enhancements to handle a variety of video quality and compression problems.
  • the following describes a number of such enhancements, most of which are preferably embodied as a set of tools which can be applied to the tasks of enhancing images and compressing such images.
  • the tools can be combined by a content developer in various ways, as desired, to optimize the visual quality and compression efficiency of a compressed data stream, particularly a layered compressed data stream.
  • Such tools include improved image filtering techniques, motion vector representation and determination, de-interlacing and noise reduction enhancements, motion analysis, imaging device characterization and correction, an enhanced 3-2 pulldown system, frame rate methods for production, a modular bit rate technique, a multi-layer DCT structure, variable length coding optimization, an augmentation system for MPEG-2 and MPEG-4, and guide vectors for the spatial enhancement layer.
  • FIG. 1 is a timing diagram showing the pulldown rates for 24 fps and 36 fps material to be displayed at 60 Hz.
  • FIG. 2 is a first preferred MPEG-2 coding pattern.
  • FIG. 3 is a second preferred MPEG-2 coding pattern.
  • FIG.4 is a block diagram showing temporal layer decoding in accordance with the preferred embodiment ofthe invention.
  • FIG. 5 is a block diagram showing 60 Hz interlaced input to a converter that can output both 36 Hz and 72 Hz frames.
  • FIG. 6 is a diagram showing a "master template" for a base MPEG-2 layer at 24 or 36 Hz.
  • FIG. 7 is a diagram showing enhancement of a base resolution template using hierarchical resolution scalability utilizing MPEG-2.
  • FIG. 8 is a diagram showing the preferred layered resolution encoding process.
  • FIG. 9 is a diagram showing the preferred layered resolution decoding process.
  • FIG. 10 is a block diagram showing a combination of resolution and temporal scalable options for a decoder in accordance with the invention.
  • FIG. 11 is diagram of a base layer expanded by using a gray area and enhancement to provide picture detail.
  • FIG. 12 is a diagram ofthe relative shape, amplitudes, and lobe polarity of a preferred downsizing filter.
  • FIGS. 13A and 13B are diagrams of the relative shape, amplitudes, and lobe polarity of a pair of preferred upsizing filters for upsizing by a factor of 2.
  • FIG. 14A is a block diagram of an odd-field de-interlacer.
  • FIG. 14B is a block diagram of an even-field de-interlacer.
  • FIG. 15 is a block diagram of a frame de-interlacer using three de-interlaced fields.
  • FIG. 16 is a block diagram of an additional layered mode based upon a 2/3 base layer.
  • FIG. 17 is a diagram of one example of applying higher bit rates to modular portions of a compressed data stream.
  • FIG. 18 graphically illustrates the relationships of DCT harmonics between two resolution layers.
  • FIG. 19 graphically illustrates the similar relationships of DCT harmonics between three resolution layers.
  • FIG. 20 is a diagram showing a set of matched DCT block sizes for multiple resolution layers.
  • FIG. 21 is a diagram showing examples of splitting of motion compensation macroblocks for determining independent motion vectors.
  • FIG. 22 is a block diagram showing an augmentation system for MPEG-2 type systems.
  • FIG. 23 is a diagram showing use of motion vectors from a base layer as guide vectors for a resolution enhancement layer
  • FIGS. 24A-24E are data flow diagrams showing on example professional level enhancement mode.
  • Beat Problems Optimal presentation on a 72 or 75 Hz display will occur if a camera or simulated image is created having a motion rate equal to the display rate (72 or 75 Hz, respectively), and vice versa. Similarly, optimal motion fidelity on a 60 Hz display will result from a 60 Hz camera or simulated image.
  • Use of 72 Hz or 75 Hz generation rates with 60 Hz displays results in a 12 Hz or 15 Hz beat frequency, respectively. This beat can be removed through motion analysis, but motion analysis is expensive and inexact, often leading to visible artifacts and temporal aliasing. In the absence of motion analysis, the beat frequency dominates the perceived display rate, making the 12 or 15 Hz beat appear to provide less accurate motion than even 24 Hz.
  • 24 Hz forms a natural temporal common denominator between 60 and 72 Hz.
  • 75 Hz has a slightly higher 15 Hz beat with 60 Hz, its motion is still not as smooth as 24 Hz, and there is no integral relationship between 75 Hz and 24 Hz unless the 24 Hz rate is increased to 25 Hz.
  • movies are often played 4% fast at 25 Hz; this can be done to make film presentable on 75 Hz displays.
  • the 3-2 pulldown pattern repeats a first frame (or field) 3 times, then the next frame 2 times, then the next frame 3 times, then the next frame 2 times, etc. This is how 24 fps film is presented on television at 60 Hz (actually, 59.94 Hz for NTSC color). That is, each of 12 pairs of 2 frames in one second of film is displayed 5 times, giving 60 images per second.
  • the 3-2 pulldown pattern is shown in FIG. 1.
  • Motion Blur m order to further explore the issue of finding a common temporal rate higher than 24 Hz, it is useful to mention motion blur in the capture of moving images.
  • Camera sensors and motion picture film are open to sensing a moving image for a portion of the duration of each frame.
  • the duration of this exposure is adjustable.
  • Film cameras require a period of time tp advance the film, and are usually limited to being open only about 210 out of 360 degrees, or a 58% duty cycle.
  • some portion ofthe frame time is often required to "read” the image from the sensor. This can vary from 10% to 50% ofthe frame time.
  • an electronic shutter must be used to blank the light during this readout time.
  • the "duty cycle" of CCD sensors usually varies from 50 to 90%, and is adjustable in some cameras. The light shutter can sometimes be adjusted to further reduce the duty cycle, if desired.
  • the most common sensor duty cycle duration is 50%.
  • Preferred Rate With this issue in mind, one can consider the use of only some of the frames from an image sequence captured at 60, 72, or 75 Hz. Utilizing one frame in two, three, four, etc., the subrates shown in TABLE 1 can be derived.
  • the rate of 15 Hz is a unifying rate between 60 and 75 Hz.
  • the rate of 12 Hz is a unifying rate between 60 and 72 Hz.
  • the desire for a rate above 24 Hz eliminates these rates.24 Hz is not common, but the use of 3-2 pulldown has come to be accepted by the industry for presentation on 60 Hz displays.
  • the only candidate rates are therefore 30, 36, and 37.5 Hz. Since 30 Hz has a 7.5 Hz beat with 75 Hz, and a 6 Hz beat with 72 Hz, it is not useful as a candidate.
  • the motion rates of 36 and 37.5 Hz become prime candidates for smoother motion than 24 Hz material when presented on 60 and 72/75 Hz displays. Both of these rates are about 50% faster and smoother than 24 Hz.
  • the rate of 37.5 Hz is not suitable for use with either 60 or 72 Hz, so it must be eliminated, leaving only 36 Hz as having the desired temporal rate characteristics.
  • the motion rate of 37.5 Hz could be used if the 60 Hz display rate for television can be move 4% to 62.5 Hz. Given the interests behind 60 Hz, 62.5 Hz appears unlikely - there are even those who propose the very obsolete 59.94 Hz rate for new television systems. However, if such a change were to be made, the other aspects ofthe invention could be applied to the 37.5 Hz rate.
  • the rates of 24, 36, 60, and 72 Hz are left as candidates for a temporal rate family.
  • the rates of 72 and 60 Hz cannot be used for a distribution rate, since motion is less smooth when converting between these two rates than if 24 Hz is used as the distribution rate, as described above.
  • 36 Hz is the prime candidate for a master, unifying motion capture and image distribution rate for use with 60 and 72/75 Hz displays.
  • the 3-2 pulldown pattern for 24 Hz material repeats a first frame (or field) 3 times, then the next frame 2 times, then the next frame 3 times, then the next frame 2 times, etc.
  • each pattern optimally should be repeated in a 2-1-2 pattern. This can be seen in TABLE 2 and graphically in FIG. 1.
  • 36 Hz is the optimum rate for a master, unifying motion capture and image distribution rate for use with 60 and 72 Hz displays, yielding smoother motion than 24 Hz material presented on such displays.
  • 36 Hz meets the goals set forth above, it is not the only suitable capture rate. Since 36 Hz cannot be simply extracted from 60 Hz, 60 Hz does not provide a suitable rate for capture. However, 72 Hz can be used for capture, with every other frame then used as the basis for 36 Hz distribution. The motion blur from using every other frame of 72 Hz material will be half of the motion blur at 36 Hz capture. Tests of motion blur appearance of every third frame from 72 Hz show that staccato strobing at 24 Hz is objectionable. However, utilizing every other frame from 72 Hz for 36 Hz display is not objectionable to the eye compared to 36 Hz native capture.
  • 72 Hz camera to achieve a 36 Hz distribution rate can profit from an increased motion blur duty cycle.
  • the normal 50% duty cycle at 72 Hz, yielding a 25% duty cycle at 36 Hz has been demonstrated to be acceptable, and represents a significant improvement over 24 Hz on 60 Hz and 72 Hz displays.
  • the duty cycle is increased to be in the 75-90% range, then the 36 Hz samples would begin to approach the more common 50% duty cycle.
  • Increasing the duty rate may be accomplished, for example, by using "backing store" CCD designs which have a short blanking time, yielding a high duty cycle. Other methods may be used, including dual CCD multiplexed designs.
  • Modified MPEG-2 Compression For efficient storage and distribution, digital source material having the preferred temporal rate of 36 Hz should be compressed.
  • the preferred form of compression for the invention is accomplished by using a novel variation ofthe MPEG-2 standard, but may be used with any other compression system with similar characteristics (e.g., MPEG-4).
  • MPEG-2 Basics.
  • MPEG-2 is an international video compression standard defining a video syntax that provides an efficient way to represent image sequences in the form of more compact coded data.
  • the language ofthe coded bits is the "syntax.” For example, a few tokens can represent an entire block of 64 samples.
  • MPEG also describes a decoding (reconstruction) process where the coded bits are mapped from the compact representation into the original, "raw" format ofthe image sequence. For example, a flag in the coded bitstream signals whether the following bits are to be decoded with a discrete cosine transform (DCT) algorithm or with a prediction algorithm.
  • DCT discrete cosine transform
  • the algorithms comprising the decoding process are regulated by the semantics defined by MPEG.
  • MPEG-2 defines a programming language as well as a data format.
  • An MPEG-2 decoder must be able to parse and decode an incoming data stream, but so long as the data stream complies with the MPEG-2 syntax, a wide variety of possible data structures and compression techniques can be used.
  • the invention takes advantage of this flexibility by devising a novel means and method for temporal and resolution scaling using the MPEG-2 standard.
  • MPEG-2 uses an intraframe and an interframe method of compression. In most video scenes, the background* remains relatively stable while action takes place in the foreground. The background may move, but a great deal of the scene is redundant. MPEG-2 starts its compression by creating a reference frame called an I (for Intra) frame. I frames are compressed without reference to other frames and thus contain an entire frame of video information. I frames provide entry points into a data bitstream for random access, but can only be moderately compressed. Typically, the data representing I frames is placed in the bitstream every 10 to 15 frames. Thereafter, since only a small portion of the frames that fall between the reference I frames are different from the bracketing I frames, only the differences are captured, compressed and stored.
  • I for Intra
  • P frames generally are encoded with reference to a past frame (either an I frame or a previous P frame), and, in general, will be used as a reference for future P frames.
  • P frames receive a fairly high amount of compression.
  • B frames pictures provide the highest amount of compression but generally require both a past and a future reference in order to be encoded.
  • Bi-directional frames are never used for reference frames.
  • Macroblocks within P frames may also be individually encoded using intra-frame coding.
  • Macroblocks within B frames may also be individually encoded using intra-frame coding, forward predicted coding, backward predicted coding, or both forward and backward, or bi-directionally interpolated, predicted coding.
  • a macroblock is a 16x16 pixel grouping of four 8x8 DCT blocks, together with one motion vector for P frames, and one or two motion vectors for B frames.
  • an MPEG data bitstream comprises a sequence of I, P, and B frames.
  • a sequence may consist of almost any pattern of I, P, and B frames (there are a few minor semantic restrictions on their placement). However, it is common in industrial practice to have a fixed pattern (e.g., IBBPBBPBBPBBPBB).
  • an MPEG-2 data stream is created comprising a base layer, at least one optional temporal enhancement layer, and an optional resolution enhancement layer. Each of these layers will be described in detail.
  • the base layer is used to carry 36 Hz source material.
  • one of two MPEG-2 frame sequences can be used for the base layer: IBPBPBP or IPPPPPP.
  • IBPBPBPBP IBPBPBP
  • IPPPPPP IPPPPPP
  • 72 Hz Temporal Enhancement Layer When using MPEG-2 compression, it is possible to embed a 36 Hz temporal enhancement layer as B frames within the MPEG-2 sequence for the 36 Hz base layer if the P frame distance is even. This allows the single data stream to support both 36 Hz display and 72 Hz display. For example, both layers could be decoded to generate a 72 Hz signal for computer monitors, while only the base layer might be decoded and converted to generate a 60 Hz signal for television.
  • IPBBBPBBBPBBBP or PBPBPBPB both allow placing alternate frames in a separate stream containing only temporal enhancement B frames to take 36 Hz to 72 Hz.
  • These coding patterns are shown in FIG.S 2 and 3, respectively.
  • the 2-Frame P spacing coding pattern of FIG. 3 has the added advantage that the 36 Hz decoder would only need to decode P frames, reducing the required memory bandwidth if 24 Hz movies were also decoded without B frames.
  • FIG. 3 is a block diagram showing that a 36 Hz base layer MPEG-2 decoder 50 simply decodes the P frames to generate 36 Hz output, which may then be readily converted to either 60 Hz or 72 Hz display.
  • An optional second decoder 52 simply decodes the B frames to generate a second 36 Hz output, which when combined with the 36 Hz output ofthe base layer decoder 50 results in a 72 Hz output (a method for combining is discussed below).
  • one fast MPEG-2 decoder 50 could decode both the P frames for the base layer and the B frames for the enhancement layer.
  • Optimal Master Format A number of companies are building MPEG-2 decoding chips which operate at around 11 MPixels/second. The MPEG-2 standard has defined some "profiles" for resolutions and frame rates. Although these profiles are strongly biased toward computer-incompatible format parameters such as 60 Hz, non-square pixels, and interlace, many chip manufacturers appear to be developing decoder chips which operate at the "main profile, main level".
  • This profile is defined to be any horizontal resolution up to 720 pixels, any vertical resolution up to 576 lines at up to 25 Hz, and any frame rate of up to 480 lines at up to 30 Hz.
  • a wide range of data rates from approximately 1.5 Mbits/second to about 10 Mbits/second is also specified.
  • the main-level, main-profile pixel rate is about 10.5 MPixels/second.
  • MPEG-2 decoder chips Although there is variation among chip manufacturers, most MPEG-2 decoder chips will in fact operate at up to 13 MPixels/second, given fast support memory. Some decoder chips will go as fast as 20 MPixels/second or more. Given that CPU chips tend to gain 50% improvement or more each year at a given cost, one can expect some near-term flexibility in the pixel rate of MPEG-2 decoder chips.
  • TABLE 4 illustrates some desirable resolutions and frame rates, and their corresponding pixel rates.
  • All of these formats can be utilized with MPEG-2 decoder chips that can generate at least 12.6 MPixels/second.
  • the very desirable 640x480 at 36 Hz format can be achieved by nearly all current chips, since its rate is 11.1 MPixels/second.
  • a widescreen 1024x512 image can be squeezed into 680x512 using a 1.5:1 squeeze, and can be supported at 36 Hz if 12.5 MPixels/second can be handled.
  • the highly desirable square pixel widescreen template of 1024x512 can achieve 36 Hz when MPEG-2 decoder chips can process about 18.9 MPixels/second. This becomes more feasible if 24 Hz and 36 Hz material is coded only with P frames, such that B frames are only required in the 72 Hz temporal enhancement layer decoders. Decoders which use only P frames require less memory and memory bandwidth, making the goal of 19 MPixels/second more accessible.
  • the 1024x512 resolution template would most often be used with 2.35:1 and
  • the invention provides a unique way of accommodating a wide variety of aspect ratios and temporal resolution compared to the prior art. (Further discussion of a master template is set forth below).
  • the temporal enhancement layer of B frames to generate 72 Hz can be decoded using a chip with double the pixel rates specified above, or by using a second chip in parallel with additional access to the decoder memory.
  • merging can be done invisibly to the decoder chip using the MPEG-2 transport layer.
  • the MPEG-2 transport packets for two PIDs can be recognized as containing the base layer and enhancement layer, and their stream contents can both be simply passed on to a double-rate capable decoder chip, or to an appropriately configured pair of normal rate decoders.
  • the data partitioning feature allows the B frames to be marked as belonging to a different class within the MPEG-2 compressed data stream, and can therefore be flagged to be ignored by 36-Hz decoders which only support the temporal base layer rate.
  • Temporal scalability as defined by MPEG-2 video compression, is not as optimal as the simple B frame partitioning ofthe invention.
  • the MPEG-2 temporal scalability is only forward referenced from a previous P or B frame, and thus lacks the efficiency available in the B frame encoding proposed here, which is both forward and backward referenced.
  • the simple use of B frames as a temporal enhancement layer provides a simpler and more efficient temporal scalability than does the temporal scalability defined within MPEG-2.
  • this use of B frames as the mechanism for temporal scalability is fully compliant with MPEG-2.
  • the two methods of identifying these B frames as an enhancement layer, via data partitioning or alternate PID's for the B frames, are also fully compliant.
  • Temporal enhancement layer 50/60 Hz Temporal enhancement layer.
  • a 60 Hz temporal enhancement layer (which encodes a 24 Hz signal) can be added in similar fashion to the 36 Hz base layer.
  • a 60 Hz temporal enhancement layer is particular useful for encoding existing 60 Hz interlaced video material.
  • FIG. 5 is a block diagram showing 60 Hz interlaced input from cameras 60 or other sources (such as non-film video tape) 62 to a converter 64 that includes a de-interlacer function and a frame rate conversion function that can output a 36 Hz signal (36 Hz base layer only) and a 72 Hz signal (36 Hz base layer plus 36 Hz from the temporal enhancement layer).
  • NTSC NTSC
  • most NTSC signals are viewed with substantial impairment on most home televisions. Further, viewers have come to accept the temporal impairments inherent in the use of 3-2 pulldown to present film on television. Nearly all prime-time television is made on film at 24 frames per second. Thus, only sports, news, and other video-original shows need be processed in this fashion. The artifacts and losses associated with converting these shows to a 36/72 Hz format are likely to be offset by the improvements associated with high-quality de-interlacing ofthe signal.
  • enhancement layer 5 and enhancement layer should appear similar to 72 Hz origination in terms of motion blur. Accordingly, few viewers will notice the difference, except possibly as a slight improvement, when interlaced 60 Hz NTSC material is processed into a 36 Hz base layer, plus 24 Hz from the temporal enhancement layer, and displayed at 60 Hz. However, those who buy new 72 Hz digital non-interlaced televisions will notice a small improvement o when viewing NTSC, and a major improvement when viewing new material captured or originated at 72 Hz. Even the decoded 36 Hz base layer presented on 72 Hz displays will look as good as high quality digital NTSC, replacing interlace artifacts with a slower frame rate.
  • PAL video tapes are best slowed to 48 Hz prior to such conversion.
  • Live PAL requires conversion using the relatively unrelated rates of 50, 36, and 72 Hz.
  • Such converter units presently are only affordable at the source of broadcast signals, and are not presently practical at each receiving device in the home and office.
  • the process of resolution enhancement can be achieved by generating a resolution enhancement layer as an independent MPEG-2 stream and applying MPEG-2 compression to the enhancement layer.
  • This technique differs from the "spatial scalability" defined with MPEG-2, which has proven to be highly inefficient.
  • MPEG-2 0 contains all ofthe tools to construct an effective layered resolution to provide spatial scalability.
  • the preferred layered resolution encoding process ofthe invention is shown in FIG. 8.
  • the preferred decoding process ofthe invention is shown in FIG. 9.
  • an original 2kxlk image 80 is down-filtered, preferably using an optimized filter having negative lobes (see discussion of FIG. 12 below), to 112 resolution in each dimension to create a 1024x512 base layer 81.
  • the base layer 81 is then compressed according to conventional MPEG-2 algorithms, generating an MPEG-2 base layer 82 suitable for transmission.
  • full MPEG-2 motion compensation can be used during this compression step.
  • That same signal is then decompressed using conventional MPEG-2 algorithms back to a 1024x512 image 83.
  • the 1024x512 image 83 is expanded (for example, by pixel replication, or preferably by better up-filters such as spline interpolation or filters having negative lobes; see discussion of FIGS. 13A and 13B below) to a first 2kxlk enlargement 84.
  • the filtered 1024x512 base layer 81 is expanded to a second 2kxlk enlargement 85.
  • This second 2kxlk enlargement 85 is subtracted from the original 2kxlk image 80 to generate an image that represents the top octave of resolution between the original high resolution image 80 and the original base layer image 81.
  • the resulting image is optionally multiplied by a sharpness factor or weight, and then added to the difference between the original 2kxlk image 80 and the second 2kxlk enlargement 85 to generate a center-weighted 2kxlk enhancement layer source image 86.
  • This enhancement layer source image 86 is then compressed according to conventional MPEG-2 algorithms, generating a separate MPEG-2 resolution enhancement layer 87 suitable for transmission.
  • full MPEG-2 motion compensation can be used during this compression step.
  • the base layer 82 is decompressed using conventional MPEG-2 algorithms back to a 1024x512 image 90.
  • the 1024x512 image 90 is expanded to a first 2kxlk image 91.
  • the resolution enhancement layer 87 is decompressed using conventional MPEG-2 algorithms back to a second 2kxlk image 92.
  • the first 2kxlk image 91 and the second 2kxlk image 92 are then added to generate a high-resolution 2kxlk image 93. Improvements Over MPEG-2.
  • the enhancement layer is created by expanding the decoded base layer, taking the difference between the original image and the decode base layer, and compressing.
  • a compressed resolution enhancement layer may be optionally added to the base layer after decoding to create a higher resolution image in the decoder.
  • the inventive layered resolution encoding process differs from MPEG-2 spatial scalability in several ways: •
  • the enhancement layer difference picture is compressed as its own MPEG-2 data stream, with I, B, and P frames. This difference represents the major reason that resolution scalability, as proposed here, is effective, where MPEG-2 spatial scalability is ineffective.
  • the spatial scalability defined within MPEG-2 allows an upper layer to be coded as the difference between the upper layer picture and the expanded base layer, or as a motion compensated MPEG-2 data stream ofthe actual picture, or a combination of both. However, neither of these encodings is efficient.
  • the difference from the base layer could be considered as an I frame of the difference, which is inefficient compared to a motion-compensated difference picture, as in the invention.
  • the upper-layer encoding defined within MPEG-2 is also inefficient, since it is identical to a complete encoding ofthe upper layer.
  • the motion compensated encoding of the difference picture, as in the invention, is therefore substantially more efficient.
  • the MPEG-2 systems transport layer (or another similar mechanism) must be used to multiplex the base layer and enhancement layer.
  • the expansion and resolution reduction (down) filtering can be a gaussian or spline function, or a filter with negative lobes (see FIG. 12), which are more optimal than the bilinear interpolation specified in MPEG-2 spatial scalability.
  • the image aspect ratio must match between the lower and higher layers in the preferred embodiment.
  • MPEG-2 spatial scalability extensions to width and/or height are allowed. Such extensions are not allowed in the preferred embodiment due to efficiency requirements.
  • the entire area ofthe enhancement layer is not coded.
  • the area excluded from enhancement will be the border area.
  • the 2kxlk enhancement layer source image 86 in the preferred embodiment is center-weighted.
  • a fading function (such as linear weighting) is used to "feather" the enhancement layer toward the center ofthe image and away from the border edge to avoid abrupt transitions in the image.
  • any manual or automatic method of determining regions having detail which the eye will follow can be utilized to select regions which need detail, and to exclude regions where extra detail is not required. All ofthe image has detail to the level ofthe base layer, so all ofthe image is present. Only the areas of special interest benefit from the enhancement layer.
  • edges or borders ofthe frame can be excluded from enhancement, as in the center-weighted embodiment described above.
  • the MPEG-2 parameters "lower_layer_prediction_horizontal&vertical offset” parameters used as signed negative integers, combined with the "horizontal&ver- tical_subsampling_factor_m&n” values, can be used to specify the enhancement layer rectangle's overall size and placement within the expanded base layer.
  • a sharpness factor is added to the enhancement layer to offset the loss of sharpness which occurs during quantization. Care must be taken to utilize this parameter only to restore the clarity and sharpness of the original picture, and not to enhance the image.
  • the sharpness factor is the "high octave" of resolution between the original high resolution image 80 and the original base layer image 81 (after expansion). This high octave image will be quite noisy, in addition to containing the sharpness and detail of the high octave of resolution.
  • the amount that should be added depends upon the level ofthe noise in the original image. A typical weighting value is 0.25. For noisy images, no sharpness should be added, and it even may be advisable to suppress the noise in the original for the enhancement layer before compressing using conventional noise suppression techniques which preserve detail.
  • Temporal and resolution scalability are intermixed by utilizing B frames for temporal enhancement from 36 to 72 Hz in both the base and resolution enhancement layers. In this way, four possible levels of decoding performance are possible with two layers of resolution scalability, due to the options available with two levels of temporal scalability. These differences represent substantial improvements over MPEG-2 spatial and temporal scalability. However, these differences are still consistent with MPEG-2 decoder chips, although additional logic may be required in the decoder to perform the expansion and addition in the resolution enhancement decoding process shown in FIG. 9. Such
  • Optional Non-MPEG-2 Coding of the Resolution Enhancement Layer It is possible to utilize a different compression technique for the resolution enhancement layer than MPEG-2. Further, it is not necessary to utilize the same compression technology for o the resolution enliancement layer as for the base layer. For example, motion-compensated block wavelets can be utilized to match and track details with great efficiency when the difference layer is coded. Even if the most efficient position for placement of wavelets jumps around on the screen due to changing amounts of differences, it would not be noticed in the low-amplitude enhancement layer. Further, it is not necessary to cover the 5 entire image - it is only necessary to place the wavelets on details. The wavelets can have their placement guided by detail regions in the image. The placement can also be biased away from the edge.
  • a 2k Ik template can efficiently support the common widescreen aspect ratios of 1.85:1 and 2.35:1.
  • a 2kxlk template can also accommodate 1.33:1 and other aspect ratios.
  • integers especially the factor of 2 and simple fractions (3/2 & 4/3) are most efficient step sizes in resolution layering, it is also possible to use arbitrary ratios to achieve any required resolution layering.
  • using a 2048x1024 template, or something near it, provides not only a high quality digital master format, but also can provide many other convenient resolutions from a factor of two base layer (lkx512), including NTSC, the U.S. television standard.
  • digital mastering formats should be created in the frame rate ofthe film if from existing movies (i. e. , at 24 frames per second) .
  • the common use of both 3 -2 pulldown and interlace would be inappropriate for digital film masters.
  • For new digital electronic material it is hoped that the use of 60 Hz interlace will cease in the near future, and be replaced by frame rates which are more compatible with computers, such as 72 Hz, as proposed here.
  • the digital image masters should be made at whatever frame rate the images are captured, whether at 72 Hz, 60 Hz, 36 Hz, 37.5 Hz, 75 Hz, 50 Hz, or other rates.
  • a mastering format as a single digital source picture format for all electronic release formats differs from existing practices, where PAL, NTSC, letterbox, pan-and-scan, HDTV, and other masters are all generally independently made from a film original.
  • PAL, NTSC, letterbox, pan-and-scan, HDTV, and other masters are all generally independently made from a film original.
  • the use of a mastering format allows both film and digital/electronic shows to be mastered once, for release on a variety of resolutions and formats.
  • Temporal enhancement is provided by decoding B frames.
  • the resolution enhancement layer also has two temporal layers, and thus also contains B frames. For 24 fps film, the most efficient and lowest cost decoders might use only
  • decoding movies at 24 fps and decoding advanced television at 36 fps could utilize a decoder without B frame capability.
  • B frames can then be utilized between P frames to yield the higher temporal layer at 72 Hz, as shown in FIG. 3, which could be decoded by a second decoder.
  • This second decoder could also be simplified, since it would only have to decode B frames.
  • the resolution enhancement layer can add the full temporal rate of 72 Hz at high resolution by adding B frame decoding within the resolution enhancement layer.
  • FIG. 10 The combined resolution and temporal scalable options for a decoder are illustrated in FIG. 10. This example also shows an allocation ofthe proportions of an approximately 18 mbits/second data stream to achieve the spatio-temporal layered Advanced Television ofthe invention.
  • a base layer MPEG-2 1024x512 pixel data stream (comprising only P frames in the preferred embodiment) is applied to a base resolution decoder 100. Approximately 5 mbits/per sec of bandwidth is required for the P frames.
  • the base resolution decoder 100 can decode at 24 or 36 fps.
  • the output of the base resolution decoder 100 comprises low resolution, low frame rate images (1024x512 pixels at 24 or 36 Hz).
  • the B frames from the same data stream are parsed out and applied to a base resolution temporal enhancement layer decoder 102. Approximately 3 mbits/per sec of bandwidth is required for such B frames.
  • the output ofthe base resolution decoder 100 is also coupled to the temporal enhancement layer decoder 102.
  • the temporal enhancement layer decoder 102 can decode at 36 fps.
  • the combined output of the temporal enhancement layer decoder 102 comprises low resolution, high frame rate images (1024x512 pixels at 72 Hz). Also in FIG. 10, a resolution enhancement layer MPEG-22kx 1 k pixel data stream
  • the output of the base resolution decoder 100 is also coupled to the high resolution enhancement layer decoder 104.
  • the high resolution enhancement layer decoder 104 can decode at 24 or 36 fps.
  • the output of the high resolution enhancement layer decoder 104 comprises high resolution, low frame rate images (2kxlk pixels at 24 or 36 Hz).
  • the B frames from the same data stream are parsed out and applied to a high resolution temporal enhancement layer decoder 106. Approximately 4 mbits/per sec of bandwidth is required for such B frames.
  • the output ofthe high resolution enhancement layer decoder 104 is coupled to the high resolution temporal enhancement layer decoder 106.
  • the output ofthe temporal enhancement layer decoder 102 is also coupled to the high resolution temporal enhancement layer decoder 106.
  • the high resolution temporal enhancement layer decoder 106 can decode at 36 fps.
  • the combined output ofthe high resolution temporal enhancement layer decoder 106 comprises high resolution, high frame rate images (2kxlk pixels at 72 Hz).
  • MPEG-2 encoding syntax also provides efficient motion representation through the use of motion-vectors in both the base and enhancement layers. Up to some threshold of high noise and rapid image change, MPEG-2 is also efficient at coding details instead of noise within an enhancement layer through motion compensation in conjunction with DCT quantization. Above this threshold, the data bandwidth is best allocated to the base layer.
  • macroblock is a 16x16 pixel grouping of four 8x8 DCT blocks, together with one motion vector for P frames, and one or two motion vectors for B frames.
  • the bits available per macroblock for each layer are shown in TABLE 6.
  • High Temporal 18 (5+3+6+4) 123 37 overall, 30/enh. layer w/border around hi-res center for comparison:
  • the available number of bits to code each macroblock is smaller in the enhancement layer than in the base layer. This is appropriate, since it is desirable for the base layer to have as much quality as possible.
  • the motion vector requires 8 bits or so, leaving 10 to 25 bits for the macroblock type codes and for the DC and AC coefficients for all four 8x8 DCT blocks. This leaves room for only a few "strategic" AC coefficients. Thus, statistically, most of the information available for each macroblock must come from the previous frame of an enhancement layer.
  • the system described here gains its efficiency by utilizing motion compensated prediction from the previous enhancement difference frame. This is demonstrably effective in providing excellent results in temporal and resolution (spatial) layered encoding.
  • the temporal scaling and resolution scaling techniques described here work well for normal-running material at 72 frames per second using a 2k lk original source. These techniques also work well on film-based material which runs at 24 ⁇ s. At high frame rates, however, when a very noise-like image is coded, or when there are numerous shot cuts within an image stream, the enhancement layers may lose the coherence between frames which is necessary for effective coding. Such loss is easily detected, since the buffer-fullness/rate-control mechanism of a typical MPEG-2 encoder/decoder will attempt to set the quantizer to very coarse settings.
  • all of the bits normally used to encode the resolution enhancement layers can be allocated to the base layer, since the base layer will need as many bits as possible in order to code the stressful material. For example, at between about 0.5 and 0.33 MPixels per frame for the base layer, at 72 frames per second, the resultant pixel rate will be 24 to 36 MPixels/second. Applying all ofthe available bits to the base layer provides about 0.5 to 0.67 million additional bits per frame at 18.5 mbits/second, which should be sufficient to code very well, even on stressful material.
  • the base-layer quantizer is still operating at a coarse level under about 18.5 mbits/second at 36 ⁇ s, then the base layer frame rate can be dynamically reduced to 24, 18, or even 12 frames per second (which would make available between 1.5 and 4 mbits for every frame), which should be able to handle even the most pathological moving image types. Methods for changing frame rate in such circumstances are known in the art.
  • the adaptive quantization level is controlled by the output buffer fullness. At the high compression ratios involved in the resolution enhancement layer ofthe invention, this mechanism may not function optimally.
  • Various techniques can be used to optimize the allocation of data to the most appropriate image regions. The conceptually simplest technique is to perform a pre-pass of encoding over the resolution enhancement layer to gather statistics and to search out details which should be preserved. The results from the pre-pass can be used to set the adaptive quantization to optimize the preservation of detail in the resolution enhancement layer.
  • the settings can also be artificially biased to be non-uniform over the image, such that image detail is biased to allocation in the main screen regions, and away from the macroblocks at the extreme edges ofthe frame. Except for leaving an enhancement-layer border at high frame rates, none of these adjustments are required, since existing decoders function well without such improvements. However, these further improvements are available with a small extra effort in the enhancement layer encoder.
  • the invention described here achieves many highly desirable features. It has been claimed by some involved in the U.S . advanced television process that neither resolution nor temporal scalability can be achieved at high definition resolutions within the approximately 18.5 mbits/second available in terrestrial broadcast. However, the invention achieves both temporal and spatial-resolution scalability within this available data rate. It has also been claimed that 2 MPixels at high frame rates cannot be achieved without the use of interlace within the available 18.5 mbits/second data rate. However, achieves not only resolution (spatial) and temporal scalability, it can provide 2 MPixels at 72 frames per second.
  • the invention is also very robust, particularly compared to the current proposal for advanced television. This is made possible by the allocation of most or all ofthe bits to the base layer when very stressful image material is encountered. Such stressful material is by its nature both noise-like and very rapidly changing. In these circumstances, the eye cannot see detail associated with the enhancement layer of resolution. Since the bits are applied to the base layer, the reproduced frames are substantially more accurate than the currently proposed advanced television system, which uses a single constant higher resolution.
  • this aspect ofthe inventive system optimizes both perceptual and coding efficiency, while providing maximum visual impact.
  • This system provides a very clean image at a resolution and frame rate performance that had been considered by many to be impossible. It is believed that this aspect ofthe inventive system is likely to outperform the advanced television formats being proposed by at this time. In addition to this anticipated superior performance, the invention also provides the highly valuable features of temporal and resolution layering.
  • MPEG-2 While the discussion above has used MPEG-2 in its examples, these and other aspects of the invention may be carried out using other compression systems.
  • the invention will work with any comparable standard that provides I, B, and P frames or equivalents, such as MPEG-1, MPEG-2, MPEG-4, H.263, and other compression systems (including wavelets and other non-DCT systems).
  • a number of enhancements to the embodiments described above may be made to handle a variety of video quality and compression problems.
  • the following describes a number of such enhancements, most of which are preferably embodied as a set of tools which can be applied to the tasks of enhancing images and compressing such images.
  • the tools can be combined by a content developer in various ways, as desired, to optimize the visual quality and compression efficiency of a compressed data stream, particularly a layered compressed data stream.
  • a resolution enhancement layer is coded using MPEG-type (e.g., MPEG-2, MPEG-4, or comparable systems) compression is to bias a difference picture with a gray bias.
  • MPEG-type e.g., MPEG-2, MPEG-4, or comparable systems
  • the difference picture is found by subtracting an expanded and decompressed base layer from an original high resolution image. Sequences of these difference pictures are then encoded as an MPEG-type difference picture stream of frames, which operates as a normal MPEG-type picture stream. The gray bias value is removed when each difference picture is added to another image (for example, the expanded decoded base layer) to create an improved resolution result.
  • motion vectors found on this difference picture stream which is often quite noisy
  • Each difference picture represents a delta adjustment which would be needed to make a perfect encoding ofthe high resolution original.
  • the pixel difference delta values can only extend for half of the range, which is nearly always sufficient, since differences are usually quite small.
  • black (typically 0) regions can be extended at most to half gray (at 127, typically), and white (typically 255) can be extended down at most to half gray (at 128, typically).
  • the difference picture could be used to create an entire black-to-white range (of 0 to 255, typically).
  • This simple relationship can be utilized to widen the aspect ratio of a final image in addition to enhancing the resolution ofthe base layer.
  • the system can provide the gray value (128, typically) for use with a difference picture in order to add full picture detail outside ofthe extent ofthe expanded, decompressed base layer.
  • FIG. 11 is diagram of a base layer expanded by using a gray area and enhancement to provide picture detail.
  • a base layer 1100 having a narrower aspect ratio is upfiltered and then expanded in area as an expanded base region 1102.
  • the expanded base region 1102 is then "padded" with a uniform mid- gray pixel value (e.g., 128) to widen its aspect ratio or otherwise increase its size (an "additional area region” 1104).
  • An enhancement layer can then be added having a small range of possible pixel values (i.e., a difference picture) for the area that coincides with the the expanded base region 1102, but a full range of possible pixel values (e.g., ⁇ 127) over the additional area region 1104, thus providing additional actual picture information.
  • the base layer can represent a narrow or shorter (or both) image extent than the enhanced image at higher resolution.
  • the enhancement layer then contains both gray-biased image difference pictures, corresponding to the extent ofthe expanded decompressed base layer (i.e., the expanded base region 1102), as well as containing actual picture. Since the compressed enhancement layer is encoded as a standard MPEG-type picture stream, the fact that the edge region is actual picture, and the inner region is a difference picture, is not distinguished, and both are coded and carried along together in the same picture stream of frames.
  • the edge region outside the expanded decompressed base layer extent, is a normal high resolution MPEG- type encoded stream. It has an efficiency which corresponds to normal MPEG- type encoding of a high resolution picture.
  • motion vectors within the difference picture region should be constrained not to point into the border region (which contains actual picture information).
  • motion vectors for the border actual-picture region should be constrained not to point into the inner difference picture region. In this way the border actual-picture region coding and difference picture region coding will be naturally separated.
  • the quantizer and rate buffer control during encoding of these hybrid difference-plus-actual-picture image-expanding enhancement layer pictures may need special adjustment to differentiate the larger extent of signals in the border actual- picture region over the inner difference picture region.
  • this technique concerning the amount ofthe extent ofthe border actual-picture region.
  • the number of bits in proportion to the overall stream is small, but the relative efficiency for the small area is reduced because ofthe number of motion vectors which cannot find matches since such matches would be off the edge ofthe border region.
  • Another way of looking at this is that the border region has a high proportion of edge to area, unlike a usual image rectangle, which has a much lower proportion of edge to area.
  • the inner rectangular picture region typical of normal digital video as is usually coded with compression such as MPEG-2 or MPEG-4, has a high degree of matches when finding motion vectors since most ofthe area within the frame, except at the very frame edges, is usually present in the previous frame.
  • compression such as MPEG-2 or MPEG-4
  • the direction of picture coming on-screen will cause one edge to have to create picture from nothing, since the image is coming from off-screen for each frame.
  • most of a normal picture rectangle is on-screen in the previous frame, allowing the motion vectors to most often find matches.
  • the border area has a much higher percentage of off-screen mismatches in previous frames for motion compensation, since the screen outer edge, as well as the difference picture inner edge, are both "out-of-bounds" for motion vectors.
  • some loss of efficiency is inherent in this approach when considered as bits per image area (or per pixel or per macroblock, which are equivalent bits-per-area measures).
  • this relative inefficiency is a sufficiently small proportion of the overall bit rate to be acceptable. If the border is relatively large, likewise, the efficiency becomes higher, and the proportion may again be acceptable. Moderately sized borders may suffer some inefficiency during pans, but this inefficiency may be acceptable.
  • the lower resolution image may also be most naturally used on narrower screens, while the higher resolution image may be more naturally viewed on larger and wider, and/or taller screens.
  • FIG. 12 is a diagram ofthe relative shape, amplitudes, and lobe polarity of a preferred downsizing filter.
  • the down filter essentially is a center- weighted function which has been truncated to a center positive lobe 1200, a symmetric pair of adjacent (bracketing) small negative lobes 1204, and a symmetric pair of adjacent (bracketing) very small outer positive lobes 1206.
  • the absolute amplitude ofthe lobes 1200, 1202, 1204 may be adjusted as desired, so long as the relative polarity and amplitude inequality relationships shown in FIG. 12 are maintained.
  • the preferred downsizing filter When creating a base layer original (as input to the base layer compression) from a low-noise high resolution original input, the preferred downsizing filter has first negative lobes which are of a normal sine function amplitude. For clean and for high resolution input images, this normal truncated sine function works well. For lower resolutions (e.g., 1280x720, 1024x768, or 1536x768), and for noisier input pictures, a reduced first negative lobe amplitude in the filters is more optimal. A suitable amplitude in such cases is about half the truncated sine function negative lobe amplitude.
  • the small first positive lobes outside ofthe first negative lobes are also reduced to lower amplitude, typically to 1/2 to 2/3 ofthe normal sine function amplitude.
  • the affect of reducing the first negative lobes is the main issue, since the small outside positive lobes do not contribute to picture noise. Further samples outside the first positive lobes preferably are truncated to minimize ringing and other potential artifacts.
  • FIGS. 13 A and 13B are diagrams ofthe relative shape, amplitudes, and lobe polarity of a pair of preferred upsizing filters for upsizing by a factor of 2.
  • a central positive lobe 1300, 1300' is bracketed by a pair of small negative lobes 1302, 1302'.
  • An asymmetrically placed positive lobe 1304, 1304' is also required.
  • These paired upfilters could also be considered to be truncated sine filters centered on the newly created samples. For example, for a factor of two upfilter, two new samples will be created for each original sample.
  • the small adjacent negative lobes 1302, 1302' have less negative amplitude than is used in the corresponding downsizing filter (FIG. 12), or than would be used in an optimal (sine-based) upsizing filter for normal images. This is because the images being upsized are decompressed, and the compression process changes the spectral distribution. Thus, more modest negative lobes, and no additional positive lobes beyond the middle ones 1300, 1300', work better for upsizing a decompressed base layer.
  • this upsizing filter preferably is used for the base layer in both the encoder and the decoder.
  • the signal path which expands the original uncompressed base layer input image uses a gaussian upfilter rather than the upfilter described above.
  • a gaussian upfilter is used for the "high octave" of picture detail, which is determined by subtracting the expanded original base- resolution input image (without using compression) from the original picture.
  • no negative lobes are used for this particular upfiltered expansion.
  • this high octave difference signal path is typically weighted with 0.25 (or 25%) and added to the expanded decompressed base layer (using the other upfilter described above) as input to the enhancement layer compression process.
  • weights 10%, 15%, 20%, 30%, and 35% are useful for particular images when using MPEG-2.
  • Other weights may also prove useful.
  • filter weights 4-8% may be optimal when used in conjunction with other improvements described below. Accordingly, this weighting should be regarded as an adjustable parameter, depending upon the encoding system, the scenes being encoded/compressed, the particular camera (or film) being used, and the image resolution.
  • the de- interlacing techniques described below are useful as input to single-layer noninterlaced MPEG-like, as well as to the layered MPEG-like compression described above.
  • noise reduction must similarly match the needs of being an input to compression algorithms, rather than just reducing noise appearance.
  • the goal is generally to reproduce, upon decompression, no more noise than the original camera or film-grain noise. Equal noise is generally considered acceptable, after compression/decompression. Reduced noise, with equivalent sharpness and clarity with the original, is a bonus.
  • the noise reduction described below achieves these goals.
  • noise reduction can be the difference between a good looking compressed/decompressed image vs. one which is unwatchably noisy.
  • the compression process greatly amplifies noise which is above some threshold of acceptability to the compressor.
  • the use of noise-reduction pre-processing to keep noise below this threshold may be required for acceptable good quality results.
  • de-graining and/or noise-reducing filtering before layered or non-layered encoding improves the ability of the compression system to perform. While de-graining or noise-reduction is most effective on grainy or noisy images prior to compression, either process may be helpful when used in moderation even on relatively low noise or low grain pictures. Any of several known de-graining or noise-reduction algorithms may be applied. Examples are "coring", simple neighbor median filters, and softening filters.
  • noise-reduction is needed is determined by how noisy the original images are.
  • the interlace itself is a form of noise, which usually will require additional noise reduction filtering, in addition to the complex de- interlacing process described below.
  • noise processing is useful in layered and non-layered compression when noise is present above a certain level.
  • video transfers from film include film grain noise.
  • Film grain noise is caused by silver grains which couple to yellow, cyan, and magenta film dyes. Yellow affects both red and green, cyan affects both blue and green, and magenta affects both red and blue. Red is formed where yellow and magenta dye crystals overlap.
  • the red, green, and blue noise is uncorrelated. In this case, it is best to process the red, green, and blue records independently. Thus, red noise is reduced with self-red processing independently of green noise and blue noise; the same approach applies to green and blue noise. Thus, noise processing is best matched to the characteristics ofthe noise source itself. In the case of a composite image (from multiple sources), the noise may differ in characteristics over different portions ofthe image. In this situation, generic noise processing may be the only option, if noise processing is needed.
  • Re-graining and/or re-noising are relatively easy effects to add in the decoder using any of several known algorithms. For example, this can be accomplished by the addition of low pass filtered random noise of suitable amplitude.
  • the preferred compression method for interlaced source which is ultimately intended for non-interlaced display includes a step to de-interlace the interlaced source before the compression steps.
  • De-interlacing a signal after decoding in the receiver, where the signal has been compressed in the interlaced mode is both more costly and less efficient than de-interlacing prior to compression, and then sending a non-interlaced compressed signal.
  • the non-interlaced compressed signal can be either layered or non-layered (i.e., a conventional single layer compression). Experimentation has shown that filtering a single field of an interlaced source, and using that field as if it were a non-interlaced full frame, gives poor and noisy compression results.
  • a field-de-interlacer is used as the first step in the overall process to create field-frames.
  • each field is de- interlaced, creating a synthesized frame where the total number of lines in the frame is derived from the half number of lines in a field.
  • an interlaced 1080 line image will have 540 lines per even and odd field, each field representing l/60th of a second.
  • the even and odd fields of 540 lines will be interlaced to create 1080 lines for each frame, which represents l/30th of a second.
  • the de-interlacer copies each scanline without modification from a specified field (e.g., the odd fields) to a buffer that will hold some ofthe de- interlaced result.
  • the remaining intermediate scanlines (in this example, the even scanlines) for the frame are synthesized by adding half of the field line above and half ofthe field line below each newly stored line.
  • the pixel values of line 2 for a frame would each comprise 1/2 ofthe summed corresponding pixel values from each of line 1 and line 3.
  • the generation of intermediate synthesized scanlines may be done on the fly, or may be computed after all ofthe scanlines from a field are stored in a buffer. The same process is repeated for the next field, although the field types (i.e., even, odd) will be reversed.
  • FIG. 14A is a block diagram of an odd-field de-interlacer, showing that the odd lines from an odd field 1400 are simply copied to a de-interlaced odd field 1402, while the even lines are created by averaging adjacent odd lines from the original odd field together to form the even lines ofthe de-interlaced odd field 1402.
  • FIG. 14B is a block diagram of an even-field de-interlacer, showing that the even lines from an even field 1404 are simply copied to a de-interlaced even field 1406, while the odd lines are created by averaging adjacent even lines from the original even field together to form the odd lines ofthe de-interlaced even field 1406. Note that this case corresponde to "top field first"; "bottom field first” could also be considered the "even” field.
  • FIG. 15 is a block diagram showing how the pixels of each output frame are composed of 25% of the corresponding pixels from a previous de-interlaced field (field-frame) 1502, 50% ofthe corresponding pixels from a current field-frame 1504, and 25% ofthe corresponding pixels from the next field-frame 1506.
  • the new de-interlaced frame then contains much fewer interlace difference artifacts between frames than do the three field-frames of which it is composed.
  • there is a temporal smearing by adding the previous field-frame and next field-frame into a current field-frame. This temporal smearing is usually not objectionable, especially in light ofthe de-interlacing improvements which result.
  • This de-interlacing process is very beneficial as input to compression, either single layer (unlayered) or layered. It is also beneficial just as a treatment for interlaced video for presentation, viewing, or making still frames, independent of use with compression.
  • the picture from the de-interlacing process appears "clearer" than the presentation ofthe interlace directly, or ofthe de-interlaced fields.
  • a threshold test may be applied which compares the result ofthe [0.25, 0.5, 0.25] temporal filter against the corresponding pixel values of only the middle field-frame. If a middle field-frame pixel value differs more than a specified threshold amount from the value ofthe corresponding pixel from the three-field-frame temporal filter, then only the middle field-frame pixel value is used.
  • a pixel from the three-field-frame temporal filter is selected where it differs less than the threshold amount from the corresponding pixel ofthe single de-interlaced middle field-frame, and the middle field-frame pixel value is used when there is more difference than the threshold.
  • This allows fast motion to be tracked at the field rate, and smoother parts ofthe image to be filtered and smoothed by the three-field-frame temporal filter.
  • This combination has proven an effective, if not optimal, input to compression. It is also very effective for processing for direct viewing to de-interlace image material (also called line doubling in conjunction with display).
  • Rdiff R_single_field_deinterlaced minus R_three_field_deinterlaced
  • Gdiff G_single_field_deinterlaced minus G_three_field_deinterlaced
  • Bdiff B_single_field_deinterlaced minus B_three_field_deinterlaced
  • ThresholdingValue abs(Rdiff+Gdiff+Bdiff) + abs(Rdiff) + abs(Gdiff)+ abs(Bdiff)
  • ThresholdingValue is then compared to a threshold setting.
  • Typical threshold settings are in the range of 0.1 to 0.3, with 0.2 being most common.
  • smooth-filtering the three-field- frame and single-field-frame de-interlaced pictures can be used before comparing and thresholding them.
  • This smooth filtering can be accomplished simply by down filtering (e.g., down filtering by two using the preferably down filter described above), and then up filtering (e.g., using a gaussian up-filter by two.
  • This "down-up" smoothed filter can be applied to both the single-field-frame de-interlaced picture and the three-field-frame de-interlaced picture.
  • the smoothed single-field-frame and three-field-frame pictures can then compared to compute a ThresholdingValue and then thresholded to determine which picture will source each final output pixel.
  • the threshold test is used as a switch to select between the single- field-frame de-interlaced picture and the three-field-frame temporal filter combination of single-field-frame de-interlaced pictures.
  • This selection then results in an image where the pixels are from the three-field-frame de-interlacer in those areas where that image differs in small amounts (i.e., below the threshold) from the single field-frame image, and where the pixels are from the single field-frame image in those areas where the three-field-frame differed more than then the threshold amount from the single-field-frame de-interlaced pixels (after smoothing).
  • This technique has proven effective in preserving single-field fast motion details (by switching to the single-field-frame de-interlaced pixels), while smoothing large portions ofthe image (by switching to the three-field-frame de-interlaced temporal filter combination).
  • a typical blending is to create new frame by adding 33.33% (1/3) of a single middle field-frame to 66.67% (2/3) ofthe corresponding three-field-frame smoothed image. This can be done before or after threshold switching, since the result is the same either way, only affecting the smoothed three-field-frame picture.
  • the variations adjust the threshold (0.018051) a little, the factor (4.5) a little (e.g. 4.0), and the exponent (0.45) a little (e.g., 0.4).
  • the fundamental formula remains the same.
  • a matrix operation such as a RGB to/from YUV conversion, implies linear values.
  • the various types of MPEG encoding are neutral to the non-linear aspects ofthe signal, although its efficiency is effected due to the use ofthe matrix conversion RGB to/from YUV.
  • the brightness variation will be represented completely in the Luminance parameter, where full detail is provided.
  • linear vs. logarithmic vs. video issue impacts filtering.
  • small signal excursions e.g. 10% or less
  • small signal excursions e.g. 10% or less
  • a linear filter is much more effective, and produces much better image quality. Accordingly, if large excursions are to be optimally coded, transformed, or otherwise processed, it would be desirable to first convert the non-linear signal to a linear one in order to be able to apply a linear filter.
  • De-interlacing is therefore much better when each filter and summation step utilizes conversions to linear values prior to filtering or summing. This is due to the large signal excursions inherent in interlaced signals at small details ofthe image.
  • the image signals are converted back to the non-linear video digital representation.
  • the three-field-frame weighting e.g., [0.25, 0.5, 0.25] or [0.1667, 0.6666, 0.1667]
  • Other filtering and weighted sums of partial terms in noise and de-interlace filtering should also be converted to linear form for computation. Which operations warrant linear processing is determined by signal excursion, and the type of filtering.
  • Image sharpening can be appropriately computed in video or logarithmic non-linear representations, since it is self-proportional.
  • matrix processing, spatial filtering, weighted sums, and de-interlace processing should be computed using linearized digital values.
  • the single field-frame de-interlacer described above computes missing alternate lines by averaging the line above and below each actual line. This average is much more correct numerically and visually if this average is done linearly.
  • the digital values are linearized first, then averaged, and then reconverted back into the non-linear video representation.
  • a 1280x720 enhancement layer can utilize an 864x480 base layer (i.e., a 2/3 relationship between the enhancement and base layer).
  • FIG. 16 is a block diagram of such a mode.
  • An original image 1600 at 1280x720 is padded to 1296x720 (to be a multiple of 16) and then downsized by 2/3 to an 864x480 image 1602 (also a multiple of 16).
  • the downsizing preferably uses a normal filter or a filter having mild negative lobes.
  • this downsized image 1602 may be input to an first encoder 1604 (e.g., an MPEG-2 or MPEG-4 encoder) for direct encoding as a base layer.
  • an first encoder 1604 e.g., an MPEG-2 or MPEG-4 encoder
  • the base layer is decompressed and upsized (expanded and up-filtered) by 3/2 to a 1296x720 intermediate frame 1606.
  • the upfilter preferably has mild negative lobes.
  • This intermediate frame 1606 is subtracted from the original image 1600.
  • the 864x480 image 1602 is up-filtered by 3/2 (preferably using a gaussian filter) to 1280x720 and subtracted from the original image 1600.
  • the result is weighted (e.g., typically by 25% for MPEG-2) and added to the result ofthe subtraction ofthe intermediate frame 1606 from the original image 1600.
  • This resulting sum is cropped to a reduced size (e.g., 1152x688) and the edges feathered, resulting in a pre-compression enhancement layer frame 1608.
  • This pre- compression enhancement layer frame 1608 is applied as an input to a second encoder 1610 (e.g., an MPEG-2 or MPEG-4 encoder) for encoding as an enhancement layer.
  • a second encoder 1610 e.g., an MPEG-2 or MPEG-4 encoder
  • the efficiency and quality at 18.5 mbits/sec is approximately equivalent between "single" layered (i.e., non-layered) and a layered system using this configuration.
  • the efficiency of a 2/3 relationship between the enhancement and base layer is not as good as when using a factor of two, since the DCT coefficients are less orthogonal between the base and enhancement layers.
  • this construction is workable, and has the advantage of providing a high quality base layer (which is cheaper to decode). This is an improvement over the single layered configuration where the entire high resolution picture must be decoded (at a higher cost), when lower resolution is all that can be provided by a particular display.
  • the layered configuration also has the advantage that the enhancement sub- region is adjustable.
  • efficiency can be controlled by adjusting the size ofthe enhancement layer and the proportion ofthe total bit rate that is allocated to the base layer vs. the enhancement layer.
  • Adjustment ofthe enhancement layer size and bit proportion can be used to optimize compression performance, especially under high stress (rapid motion or many scene changes). For example, as noted above, all ofthe bits may be allocated to the base layer under extreme stress.
  • a source picture of 2048x1024 could have a base layer of 1536x512, which has a horizontal relationship of 3/4 and a vertical relationship of 1/2 with respect to the source image.
  • this is not optimal (a factor of two both horizontally and vertically is optimal), it is illustrative ofthe principle.
  • the use of 2/3 both horizontally and vertically might be improved upon for some resolutions by using a factor of 2 vertically and a factor of 2/3 horizontally.
  • the most useful filter is the median filter.
  • a three element median filter just ranks the three entries, via a simple sort, and picks the middle one. For example, an X (horizontal) median filter looks at the red value (or green or blue) of three adjacent horizontal pixels, and picks the one with the middle-most value. If two are the same, that value is selected. Similarly, a Y filter looks in the scanlines above and below the current pixel, and again picks the middle value.
  • each new pixel is the 50% equal average ofthe X and Y medians for the corresponding pixel from a source image.
  • a temporal median Another beneficial source of noise reduction is information from the previous and subsequent frame (i.e., a temporal median).
  • motion analysis provides the best match for moving regions.
  • it is compute intensive If a region ofthe image is not moving, or is moving slowly, the red values (and green and blue) from a current pixel can be median filtered with the red value at that same pixel location in the previous and subsequent frames.
  • odd artifacts may occur if significant motion is present and such a temporal filter is used.
  • a threshold be taken first, to determine whether such a median would differ more than a selected amount from the value of a current pixel.
  • the threshold can be computed essentially the same as for the de-interlacing threshold above:
  • Rdiff R_current_pixel minus R_temporal_median
  • Gdiff G_current_pixel minus G_temporal_median
  • ThresholdingValue abs(Rdiff+Gdiff+Bdiff) + abs(Rdiff) + abs(Gdiff)+ abs(Bdiff)
  • the ThresholdingValue is then compared to a threshold setting. Typical threshold settings are in the range 0.1 to 0.3, with 0.2 being typical. Above the threshold, the current value is kept. Below the threshold, the temporal median is used.
  • An additional median type is a median taken between the X, Y, and temporal medians.
  • Another median type can take the temporal median, and then take the equal average ofthe X and Y medians from it.
  • a preferred combination of medians is a linear weighted sum (see the discussion above on linear video processing) of five terms to determine the value for each pixel of a current image: 50% ofthe original image (thus, the most noise reduction is 3db, or half);
  • de-interlacing and noise reduction can also be improved by use of motion analysis. Adding the pixels at the same location in three fields or three frames is valid for stationary objects. However, for moving objects, if temporal averaging/smoothing is desired, it is often more optimal to attempt to analyze prevailing motion over a small group of pixels. For example, an nxn block of pixels (e.g., 2x2, 3x3, 4x4, 6x6, or 8x8) can be used to search in previous and subsequent fields or frames to attempt to find a match (in the same way MPEG-2 motion vectors are found by matching 16x16 macroblocks).
  • nxn block of pixels e.g., 2x2, 3x3, 4x4, 6x6, or 8x8
  • a "trajectory" and "moving mini-picture” can be determined.
  • the motion analysis preferably is performed by comparison of an nxn block in the current thresholded de-interlaced image with all nearby blocks in the previous and subsequent one or more frames.
  • the comparison may be the absolute value of differences in luminance or RGB over the nxn block.
  • One frame is sufficient forward and backward if the motion vectors are nearly equal and opposite. However, if the motion vectors are not nearly equal and opposite, then an additional one or two frames forward and backward can help determine the actual trajectory.
  • different de- interlacing treatments may be useful in helping determine the "best guess" motion vectors going forward and back.
  • One de-interlacing treatment can be to use only individual de-interlaced fields, although this is heavily prone to aliasing and artifacts on small moving details.
  • Another de-interlacing technique is to use only the three- field-frame smooth de-interlacing, without thresholding, having weightings [0.25, 0.5, 0.25], as described above. Although details are smoothed and sometimes lost, the trajectory may often be more correct.
  • a "smoothed nxn block" can be created by temporally filtering using the motion-vector-offset pixels from the one (or more) previous and subsequent frames.
  • a typical filter might again be [0.25, 0.5, 0.25] or [0.1667, 0.6666, 0.1667] for three frames, and possibly [0.1, 0.2, 0.4, 0.2, 0.1] for two frames back and forward.
  • filter weights can be applied to: individual de-interlaced motion- compensated field-frames; thresholded three-field-frame de-interlaced pictures, described above; and non-thresholded three-field-frame de-interlaced images, with a [0.25, 0.5, 0.25] weighting, also as described above.
  • the best filter weights usually come from applying the motion-compensated block linear filtering to the thresholded three-field-frame result described above. This is because the thresholded three-field-frame image is both the smoothest (in terms of removing aliasing in smooth areas), as well as the most motion-responsive (in terms of defaulting to a single de-interlaced field-frame above the threshold).
  • the motion vectors from motion analysis can be used as the inputs to multi-frame or multi-de-interlaced-field- frame or single-de-interlaced field-frame filters, or combinations thereof.
  • the thresholded multi-field-frame de-interlaced images form the best filter input in most cases.
  • the use of motion analysis is computationally expensive for a large search region, when fast motion might be found (such as ⁇ 32 pixels). Accordingly, it may be best to augment the speed by using special-purpose hardware or a digital signal processor assisted computer.
  • motion vectors Once motion vectors are found, together with their absolute difference measure of accuracy, they can be utilized for the complex process of attempting frame rate conversion.
  • occlusion issues objects obscuring or revealing others
  • Occlusion can also involve temporal aliasing, as can normal image temporal undersampling and its beat with natural image frequencies (such as the "backward wagon wheel” effect in movies).
  • temporal aliasing can also involve temporal aliasing, as can normal image temporal undersampling and its beat with natural image frequencies (such as the "backward wagon wheel” effect in movies).
  • De-interlacing is a simple form ofthe same problem. Just as with frame-rate- conversion, the task of de-interlacing is theoretically impossible to perform perfectly. This is especially due to the temporal undersampling (closed shutter), and an inappropriate temporal sample filter (i.e., a box filter). However, even with correct samples, issues such as occlusion and interlace aliasing further ensure the theoretical impossibility of correct results. The cases where this is visible are mitigated by the depth ofthe tools, as described here, which are applied to the problem. Pathological cases will always exist in real image sequences. The goal can only be to reduce the frequency and level of impairment when these sequences are encountered. However, in many cases, the de-interlacing process can be acceptably fully automated, and can run unassisted in real-time. Even so, there are many parameters which can often benefit from manual adjustment.
  • the filter parameters for the median filtering described above for an original image should be matched to the noise characteristics ofthe film grain or image sensor that captured the image. After this median filtered image is down-filtered to generate an input to the base layer compression process, it still contains a small amount of noise. This noise may be further reduced by a combination of another X-Y median filters (equally averaging the X and Y medians), plus a very small amount ofthe high frequency smoothing filter.
  • a preferred filter weighting of these three terms, applied to each pixel ofthe base layer, is: 70% ofthe original base layer (down filtered from median-filtered original above);
  • This small amount of additional filtering in the base layer provides a small additional amount of noise reduction and improved stability, resulting in better MPEG encoding and limiting the amount of noise added by such encoding.
  • MPEG-4 reference filters have been implemented for shifting macroblocks when finding the best motion vector match, and then using the matched region for motion compensation.
  • MPEG-4 video coding like MPEG-2, supports 1/2 pixel resolution of motion vectors for macroblocks. Unlike MPEG-2, MPEG-4 also supports 1/4 pixel accuracy.
  • the filters used are sub-optimal. In MPEG-2, the half-way point between pixels is just the average ofthe two neighbors, which is a sub-optimal box filter. In MPEG-4, this filter is used for 1/2 pixel resolution.
  • effects of filtering are significantly improved by using a negative lobe truncated sine function for filtering the 1/8-pixel points for U and V chrominance when using 1/4 pixel luminance resolution, and by using 1/4 pixel resolution filters with similar negative lobe filters when using 1/2 pixel luminance resolution.
  • These filters may be applied to video images under MPEG-1, MPEG-2, MPEG-4 or any other appropriate motion-compensated block-based image coding system.
  • each individual electronic camera, each camera type, each film- type, and each individual film scanner and scanner type used in creating input to a compression/decompression system should be individually characterized in terms of color alignment and noise (electronic noise for video cameras and scanners, and grain for film).
  • the information about where the image was created, a table ofthe specific properties, and specific settings of each piece of equipment, should be carried with the original image, and subsequently used in pre-processing prior to compression.
  • a specific camera may require a color realignment. It may also be set on a medium noise setting (substantially affecting the amount of noise processing needed). These camera settings and inherent camera properties should be carried as side information along with each shot from that camera. This information then can be used to control the type of pre-processing, and the settings of parameters for the pre- processes. For images which are edited from multiple cameras, or even composited from multiple cameras and/or film sources, the pre-processing should probably be performed prior to such editing and combining. Such pre-processing should not degrade image quality, and may even be invisible to the eye, but does have a major impact on the quality ofthe compression.
  • GB preferably expressed in pixel units.
  • the 3-2 pulldown method is used because 24 frames per second do not divide evenly into 59.94 or 60 fields per second for existing NTSC (and some proposed HDTV) systems.
  • the odd frames (or even) are placed on two ofthe interlaced fields, and the even frames (or odd) are placed on three ofthe interlaced fields.
  • one field is a duplicate in every five fields.
  • One frame of film maps to five fields of video.
  • this process leads to numerous unpleasant problems.
  • Most video processing equipment only applies its process to an immediate signal. With this being the case, a time-changing effect will operate differently on one field than the next, even though some ofthe input fields were duplicates.
  • Telecine i.e., convert from film to video
  • all original images with a deterministic cadence (i.e., either always 3 then 2, or 2 then 3) if the telecine does not provide direct 24 ⁇ s output.
  • a deterministic cadence i.e., either always 3 then 2, or 2 then 3
  • any particular processing device requiring 3-2 pulldown as input and ou ⁇ ut will get its input(s) made on the fly in real time from a 24 ⁇ s source.
  • the cadence will always begin in a standard way for each input.
  • the cadence ofthe device's output is then known, and must be identical to the cadence created on the fly as the devices' input.
  • the cadence is then un-done by this a priori knowledge, and the frames are saved in 24 ⁇ s format on the storage medium.
  • This methodology requires real-time 3-2 pulldown undo and 3-2 pulldown synthesis. Unless the cadence comes from tape in an unknown format, the 24 ⁇ s nature ofthe frames will automatically be preserved by such a film-based telecine post-production system. The system will then automatically form an optimal input to compression systems (including layered compression process described above).
  • This process should be broadly useful in video and HDTV telecine facilities. Someday, when all devices accept a 24 ⁇ s (and other rate progressive scan) native signal input, ou ⁇ ut, processing, and storage modes, such a methodology will no longer be needed. However, in the interim, many devices require 3-2 pulldown for the interface in and out, even though the devices have a targeted function to operate on film input. During this interim, the above methodology eliminates 3-2 pulldown problems and can be an essential element ofthe efficiency of post-production and telecine of film.
  • center frame helps strike a balance between the clarity of a single frame, due to the short motion blur, plus the needed blur from the adjacent frames in order to help smooth the stutter of 24 ⁇ s motion (by simulating 24 ⁇ s motion blur).
  • This weighting technique works well in about 95% of all cases, allowing this simple weighting function to provide the majority ofthe 24 ⁇ s conversions. For the remaining 5% or so ofthe cases, motion compensation can be used, as taught in U.S. patent application Serial No. 09/435,277. By having reduced the workload on the conversion process by a factor of 20 by this simple weighting technique, the remaining motion-compensated conversions become more practical when needed. It should also be noted that a 120 ⁇ s source can be used with five weightings to achieve a similar results at 24 ⁇ s. For example, weightings of [0.1, 0.2, 0.4, 0.2, 0.1] may be used.
  • 60 ⁇ s can be derived from 120 ⁇ s by taking every other frame, although the shorter open shutter duration will be noticeable on fast motion.
  • an overlapping filter can also be used (e.g., preferably about [0.1667, 0.6666, 0.1667], but may be in the range [0.1, 0.8, 0.1] to [0.25, 0.5, 0.25]), repeating the low-amplitude weighted frames.
  • frame rates allow even more careful shaping ofthe temporal sample for deriving 24 ⁇ s and other frame rates. As the frame rates become very high, the techniques of U.S. Patent No.
  • the "blocking" (setting up) ofthe shots in a scene can be checked to ensure that the 24 ⁇ s results will look good (in addition to the 72 ⁇ s or other higher rate full-rate versions).
  • the benefits of high frame rate capture are fully integrated with the capability to provide 24 ⁇ s international film and video release.
  • bit rate it is useful in many video compression applications to "modularize" the bit rate.
  • Variable bit rate systems have used continuously varying bit rates to attempt to apply more bits to faster changing shots. This can be done in a coarse way by giving each useful unit a different bit rate.
  • suitable units include a range of frames (a "Group of Pictures,” or GOP) or each P frame.
  • GOP Group of Pictures
  • the bit rate might be constant within a GOP.
  • a higher constant bit rate 5 can be utilized.
  • This is similar to the above-described layering technique of applying all ofthe bits in an enhancement layer to the base layer during periods of high stress (typically resetting at the next I frame).
  • more bits can be applied to single layer compressions, or to the base and enhancement layer (in the case of layered compression), so as to yield o high quality during periods of high stress.
  • FIG. 17 is a diagram of one example of applying higher bit rates to modular portions of a compressed data stream.
  • Groups of pictures containing normal 0 scenes 1800, 1802 are allocated bits at a constant rate.
  • a GOP 1804 occurs that . contains a scene exhibiting a high level of stress (i.e., changes that are difficult for the compression process to compress as much as "normal" scenes)
  • a higher number of bits e.g., 50-100% additional
  • MPEG-2 implementations use a constant bit rate. Constant bit rate provides a good match with constant bit rate transport and storage media. Transport systems such as broadcast channels, satellite channels, cables, and fibers, all have a fixed constant total capacity. Also, digital compressed video tape storage systems have a constant tape playback rate, thereby yielding a constant recording or playback bit rate.
  • Other MPEG-2 implementations such as DirecTV/DSS, and DVD, use some form of variable bit rate allocation. In the case of DirecTV/DSS, the variability is a combination of scene stress in the current program vs. scene stress in adjacent TV programs which share a common multiplex. The multiplex corresponds to a tuned satellite channel and transponder, which has a fixed total bit rate.
  • the digital optical disk capacity is 2.5 Gbytes, requiring that the MPEG-2 bit rate average 4.5 mbits/s for a two-hour movie.
  • the optical disk has a peak reading rate capability of 100% higher, at 9 mbits/s.
  • the average rate can be higher, up to the full 9mbits/s.
  • the way that the bit rate achieves an average of 4.5mbits/s is that a rate above this is used for scenes having high scene stress (high change due to rapid scene motion), while a rate below this average is used during low scene stress (low change due to little motion).
  • bit rate in MPEG-2 and MPEG-4 is held constant by a combination of modeling of a virtual decoder buffer's capacity, and by varying the quantization parameter to throttle the bit rate emitted from the encoder.
  • a constant quantization parameter will yield a variable number of bits, in proportion to scene change and detail, also known as scene "entropy".
  • a constant quantization parameter yields relatively constant quality, but variable bit rate.
  • a varying quantization parameter can be used in conjunction with a size bounded decoder buffer to smooth out any variability and provide a constant bit rate.
  • the sharing of many channels in a multiplex is one method that can support variable bit rate, as with DirecTV, or with standard definition signals in the ACATS/ATSC 19.3 mbits/s 6Mhz multiplex.
  • variable bit rate systems have a peak bit rate, usually somewhere near 100% above the average. Thus, these systems become constant bit rate systems at the highest bit rate, limiting the peak bit rate available for periods of continued high scene stress. There is also a limit to the input bit rate in some MPEG-2 decoder systems, also limiting the peak bit rate in such variable bit rate systems. However, this limit on peak input bit rate is gradually rising well above these other limits, as decoders improve.
  • each of these prior bit rate control systems is that there is a small memory buffer in the decoder, holding somewhere between a fraction of a frame and a few frames of moving image.
  • this decoder bit rate buffer was conceived, around 1990, there was concern that the memory cost of this buffer in decoders would have a significant affect on the decoder's price.
  • the cost of this buffer has proven insignificant. In fact, many seconds' worth of buffer is now an insignificant cost. It may be extrapolated that, in the near future, the bit receiving memory buffer may hold many minutes of video information at insignificant cost. Further, the cost of disk and other storage media has also fallen rapidly, while capacity has increased rapidly. It is therefore also reasonable to spool the compressed bitstream to disk or other storage memory systems, thereby yielding many hours or days worth of storage capacity. This is currently being done by commercially available harddrive based home video recorders.
  • FIG. 18 graphically illustrates the relationships of DCT harmonics between two resolution layers.
  • the base layer utilizes DCT coefficients using an arithmetic harmonic series having frequencies of 1, 2, 3, 4, 5, 6, and 7 times the 8x8 pixel DCT block size 1900.
  • these base layer harmonics then map to frequencies of 1/2, 1, 3/2, 2, 5/2, 3, and 7/2 ofthe corresponding enhancement layer DCT block 1902.
  • frequencies of 2, 4, and 6 times the macroblock size from the base layer are aligned with frequencies of 1, 2, and 3 times the macroblock size from the enhancement layer.
  • SNR signal-to-noise ratio
  • the 3, 5, and 7 terms from the base layer are non- harmonic with the enhancement layer, and therefore represent orthogonality to the base layer only, providing no synergy with the enhancement layer.
  • the remaining terms in the enhancement layer, 4, 5, 6, and 7, represent additional detail which the enhancement layer can provide to the image, without overlap with the base layer.
  • FIG. 19 graphically illustrates the similar relationships of DCT harmonics between three resolution layers, showing a highest enhancement layer 1904.
  • a solution to providing cross-layer orthogonality is to utilize different DCT block sizes for each resolution layer. For example, if a given layer doubles the resolution, then the DCT block size will be twice as large. This results in a harmonically aligned resolution layering structure, providing optimal coding efficiency due to optimal inter-layer coefficient orthogonality.
  • FIG. 20 is a diagram showing various DCT block sizes for different resolution layers.
  • a 4x4 pixel DCT block 2000 could be used at the base layer
  • an 8x8 pixel DCT block 2002 could be utilized at the next layer up
  • a 16x16 pixel DCT block 2004 could be utilized at the third layer
  • a 32x32 pixel DCT block 2006 could be utilized at the fourth layer.
  • each layer adds additional harmonic terms in full orthogonality to the layer(s) below.
  • additional precision in the SNR sense
  • the 16x16 pixel subset 2008 in the 32x32 pixel block 2006 can be used to augment (in an SNR improvement sense) the precision ofthe 16x16 pixel DCT block 2004.
  • macroblocks corresponding to motion vectors consist of 16x16 pixels, organized as four 8x8 DCT blocks.
  • each macroblock can optionally be further subdivided into 8x8 regions, corresponding to the DCT blocks, each having their own motion vector.
  • the motion compensation macroblocks need not be constrained by this structure.
  • the simplest structure is where the single motion vector for each base layer motion compensation macroblock applies to all higher layers as well, eliminating motion vectors from all enhancement layers entirely, since the motion is specified by the base layer's motion vector for all layers together.
  • a more efficient structure is to allow each layer to independently select (1) no motion vector (i.e., use the base layer motion vector), (2) additional sub-pixel precision for the base layer's motion vector, or (3) split each motion compensation macroblock into two, four, or other numbers of blocks each having independent motion vectors.
  • OBMC overlapped block motion compensation
  • each DCT block at each layer may be split into as many motion vector blocks for motion compensation as are optimal for that layer.
  • FIG. 21 is a diagram showing examples of splitting of motion compensation macroblocks for determining independent motion vectors.
  • the base layer if constructed using 4x4 pixel DCT blocks 2100, could utilize from one (shown) to as many as 16 motion vectors (one for each pixel), or even utilize sub-pixel motion vectors.
  • each higher level can split its larger corresponding DCT block 2102, 2104, 2106 as appropriate, yielding an optimal balance between coding prediction quality (thus saving DCT coefficient bits) vs. the bits required to specify the motion vectors.
  • the block split for motion compensation is a tradeoff between the bits used to code the motion vectors and the improvement in picture prediction.
  • variable length codes such as Huffman or arithmetic codes
  • MPEG-1, MPEG-2, MPEG-4, H.263, and other compression systems are selected based upon demonstrated efficiency on a small group of test sequences. These test sequences are limited in the types of images, and only represent a relatively narrow range of bit rate, resolution, and frame rate. Further, the variable length codes are selected based upon average performance over each test sequence, and over the test sequences as a group.
  • variable length coding system can be obtained by (1) applying specific variable length coding tables to each frame and (2) selecting the most optimal codes for that particular frame.
  • Such a selection of optimal variable length codes can applied in units smaller than a frame (a part or region of a frame), or in groups of several frames.
  • the variable length codes used for motion vectors, DCT coefficients, macroblock types, etc. can then each be independently optimized for the instantaneous conditions of a given unit (i.e., frame, sub-frame, or group of frames) at that unit's current resolution and bit rate.
  • This technique is also applicable to the spatial resolution enhancement layers described in other parts of this description.
  • variable length codes The selection of which group of variable length codes is to be used can be conveyed with each frame (or subpart or group) using a small number of bits. Further, custom coding tables can be downloaded where reliable data transmission or playback is available (such as with optical data disk or optical fiber networks).
  • MPEG-2 capable decoders.
  • DVD players and DirecTV satellite receivers are now in millions of homes.
  • the improvement which MPEG-4 video compression coding could offer beyond MPEG-2 is not yet available, since MPEG-4 is incompatible with MPEG-2.
  • MPEG-4 and MPEG-2 are both motion-compensated DCT compression system, sharing a common basic structure.
  • the composition system in MPEG-4's video coding system is fundamentally different from MPEG-2, as are some other expanded features. In this discussion, only the full frame video coding aspects of MPEG-4 are being considered.
  • MPEG-4 can optionally split a 16x16 macroblock into four 8x8 blocks, one for each DCT, each having an independent motion vector.
  • MPEG-4 B-frames have a "direct” mode, which is a type of prediction.
  • MPEG-4 B-frames do not support "I” macroblocks, unlike MPEG-2 which does support "I” macroblocks in B-frames.
  • the DCT coefficients in MPEG-4 can be coded by more elaborate patterns than with MPEG-2, although the well-known zigzag pattern is common to both MPEG-2 and MPEG-4.
  • MPEG-4 supports 10-bit and 12-bit pixel depths, whereas MPEG-2 is limited to 8 bits.
  • MPEG-4 supports quarter-pixel motion vector precision, whereas MPEG-2 is limited to half-pixel precision.
  • FIG. 22 is a block diagram showing an augmentation system for MPEG-2 type systems.
  • a main compressed data stream 2200 (shown as including motion vectors, DCT coefficients, macroblock mode bits, and I, B, and P frames) is conveyed to a conventional MPEG-2 type decoder 2202 and to a parallel enhanced decoder 2204.
  • an enhanced data stream 2206 (shown as including quarter-pixel motion vector precision, 8x8 four-way block split motion vectors, and 10-bit and 12-bit pixel depths) is conveyed to the enhanced decoder 2204.
  • the enhanced decoder 2204 would combine the two data streams 2200, 2206 and decode them to provide an enhanced video output.
  • any coding enhancements can be added to any motion-compensated DCT compression system.
  • the use of this structure can be biased by an encoder toward more optimal
  • MPEG-2 decoding or toward more optimal enhanced decoding.
  • the expectation is that such enhanced decoding, by adding MPEG-4 video coding improvements, would be favored, to achieve the optimal enhanced picture quality, with a small compromise in quality to the MPEG-2 decoded picture.
  • the MPEG-2 motion vectors can be used as "predictors" for the four-way split motion vectors (in those cases where MPEG-4 chooses to split four ways), or may be used directly for non-split 16x16 macroblocks.
  • the quarter pixel motion vector resolution can be coded as one additional bit of precision (vertically and horizontally) in the enhanced data stream 2206.
  • the extra pixel depth can be coded as extra precision to the DCT coefficients prior to applying the inverse DCT function.
  • the spatial resolution layering which is a principal subject of this invention performs most optimally when the base layer is as perfectly coded as possible.
  • MPEG- 2 is an imperfect coding, yielding degraded performance for resolution enhancement layers.
  • the base layer can be improved, for example, by using the MPEG-4 improvements described above (as well as other improvements set forth in this description) to augment the MPEG-2 data stream that encodes the base layer.
  • the resulting base layer, with accompanying enhancement data stream will then have most ofthe quality and efficiency that would have been obtained using an improved base layer which would have resulted from better coding (such as with MPEG-4 and the other improvements of this invention).
  • the resulting improved base layer can then have one or more resolution enhancement layers applied, using other aspects of this invention.
  • Motion vectors comprise a large portion ofthe allocated bits within each resolution enhancement layer created in accordance with the invention. It has been determined that it is possible to substantially reduce the number of bits required for enhancement layer motion vectors by using the corresponding motion vectors at the same position in the base layer as "guide vectors".
  • the enhancement layer motion vectors are therefore coded by only searching for a small search range about the corresponding guide vector center from the base layer. This is especially important with MPEG-4 enhancement layers, since each macroblock can optionally have 4 motion vectors, and since quarter-pixel resolution of motion vectors is available.
  • FIG. 23 is a diagram showing use of motion vectors from a base layer 2300 as guide vectors for a resolution enhancement layer 2302.
  • the process is the same for all ofthe motion vectors from the base layer. For example, in MPEG-4 a 16x16 pixel base layer macroblock may optionally be split into four 8x8 pixel motion vector blocks. A corresponding factor-of-two enhancement layer would then utilize the co- located motion vectors from the base layer as guide vectors.
  • a motion vector from one ofthe 8x8 motion vector blocks in the base layer would guide the search for a motion vector in a corresponding 16x16 pixel macroblock in the enhancement layer.
  • This 16x16 block could optionally be further split into four 8x8 motion vector blocks, all using the same corresponding base layer motion vector as a guide vector.
  • This guide-vector technique is applicable to MPEG-2, MPEG-4, or other appropriate motion-compensated spatial resolution enhancement layer(s).
  • FIGS. 24A-24E are data flow diagrams showing on example professional level enhancement mode. This figures shows picture data (including intermediate stages) in the left column, processing steps in the middle column, and ou ⁇ ut in the right column. It should be noted that this is just one example of how to combine a number ofthe processing steps described herein. Different combinations, simpler as well as more complex, can be configured to achieve different levels of compression, aspect ratios, and image quality.
  • FIG. 24A shows an initial picture 2400 at 2kxlk pixels. Down filter 2402 this image to lkx512 pixels 2404. Create motion vectors 2406 from the initial picture and output as a file 2407. Compress/decompress 2408 the lkx512 pixel image 2404 to a lkx512 decompressed image 2410 and output the compressed version as the base layer 2412, along with the associated motion vector files 2416. Expand 2418 the lkx512 decompressed image 2410 as a 2kxlk image 2420. Expand 2422 the lkx512 image 2404 as a 2kxlk image 2424. Subtract 2426 the 2kxlk image 2420 from the original image 2400 to create a 2kxlk difference picture 2428.
  • Encode/decode 2442 the combined difference picture 2440 using the original motion vectors and ou ⁇ ut an encoded enhancement layer 2444 (MPEG-2, in this example), and a 2kxlk decoded enhanced layer 2246.
  • Add 2448 the 2kxlk decoded enhanced layer 2246 to the 2kxlk image 2420 to create a 2kxlk reconstructed full base plus enhancement image 2450.
  • Subtract 2452 the original image 2400 from the 2kxlk reconstructed full base plus enhancement image 2450 to create a 2kxlk second layer difference picture 2454.
  • Increase 2456 the amplitude ofthe 2kxlk second layer difference picture 2454 to create a 2kxlk difference picture 2458.
  • red channel information 2458, the green channel information 2460, and the blue channel information 2462 to create respective red difference 2464, green difference 2466, and blue difference 2468 images.
  • encode/decode 2470 a second red layer from the red difference picture 2464 as a red second enhancement layer 2472, and a decoded red difference image 2474
  • encode/decode 2476 a second green layer from the green difference picture 2466 as a green second enhancement layer 2478, and a decoded green difference image 2480
  • encode/decode 2482 a second blue layer from the blue difference picture 2468 as a blue second enhancement layer 2484, and a decoded blue difference image 2486.
  • the invention may be implemented in hardware or software, or a combination of both. However, preferably, the invention is implemented in computer programs executing on one or more programmable computers each comprising at least a processor, a data storage system (including volatile and non- volatile memory and/or storage elements), an input device, and an ou ⁇ ut device. Program code is applied to input data to perform the functions described herein and generate ou ⁇ ut information. The ou ⁇ ut information is applied to one or more ou ⁇ ut devices, in known fashion.
  • Each such program may be implemented in any desired computer language (including machine, assembly, or high level procedural, logical, or object oriented programming languages) to communicate with a computer system.
  • the language may be a compiled or interpreted language.
  • Each such computer program is preferably stored on a storage media or device
  • inventive system may also be considered to be implemented as a computer-readable storage medium, configured with a computer program, where the storage medium so configured causes a computer system to operate in a specific and predefined manner to perform the functions described herein.
  • Allowing higher bit rates for compression units (such as the GOP) during periods of high compression stress (either automatically, by detecting high values of rate control quantization parameter, or manually controlled).
  • variable bit rate systems with one or more layers ofthe present layered compression system. • Using combined fixed and variable bit rate systems used with various layers ofthe present layered compression system.
  • DCT block size Using correspondingly larger DCT block size and additional DCT coefficients for use in resolution layering (also called "spatial scalability"). For example, if a given layer doubles the resolution, then the DCT block size will be twice as large. This results in a harmonically aligned resolution layering structure, providing optimal coding efficiency due to optimal inter-layer coefficient orthogonality.
  • Using multiple motion vectors per DCT block so that both large and small DCT blocks can optimize the tradeoff between motion vector bits and improved motion compensated prediction.
  • negative-lobed upsizing and downsizing filters particularly truncated sine filters.
  • Using negative-lobed motion compensation displacement filters Using negative-lobed motion compensation displacement filters.

Abstract

A method and apparatus for image compression using temporal and resolution layering of compressed image frames (Figs. 8). In particular, layered compression (Fig. 8, items 81, 85, and 86) allows a form of modularized decomposition of an image that supports flexible application of a variety of image enhancement techniques (Fig. 2). Further, the invention provides a number of enhancements to handle a variety of video quality and compression problems (Fig. 7). Most of the enhancements are preferably embodied as a set of tools which can be applied to the tasks of enhancing image (Fig. 8.,item 82, 86) and compressing such images. The tools can be combined by a content developer in various ways, as desired, to optimize the visual quality and compression efficiency of a compressed data stream. Such tools include improved image filtering techniques, motion vector representation and determination, de-interlacing and noise reduction enhancements, motion analysis, imaging device characterization and correction, an enhanced 3-2 pulldown system (Fig. 1), frame rate methods for production, a modular bit rate technique, a multi-layer DCT structure (Fig. 20), variable length coding optimization, an augmentation system for MPEG-2 and MPEG-4, and guide vectors for the spatial enhancement layer (Fig.8, item 87).

Description

ENHANCED TEMPORAL AND RESOLUTION LAYERING IN
ADVANCED TELEVISION
CROSS-REFERENCE TO RELATED APPLICATIONS
This application is a continuation-in-part application of and claims priority to U.S. Application Serial No. 09/442,595 filed on 11/17/99, which was a continuation of U.S. Application Serial No. 09/217,151 filed on 12/21/98, which was a continuation of U.S. Application Serial No. 08/594,815 filed 1/30/96 (now U.S. Patent No. 5,852,565, issued 12/22/98).
TECHNICAL FIELD This invention relates to electronic communication systems, and more particularly to an advanced electronic television system having temporal and resolution layering of compressed image frames having enhanced compression, filtering, and display characteristics.
BACKGROUND The United States presently uses the NTSC standard for television transmissions.
However, proposals have been made to replace the NTSC standard with an Advanced Television standard. For example, it has been proposed that the U.S. adopt digital standard-definition and advanced television formats at rates of 24 Hz, 30 Hz, 60 Hz, and 60 Hz interlaced. It is apparent that these rates are intended to continue (and thus be compatible with) the existing NTSC television display rate of 60 Hz (or 59.94 Hz). It is also apparent that "3-2 pulldown" is intended for display on 60 Hz displays when presenting movies, which have a temporal rate of 24 frames er secon (fps). However, while the above proposal provides a menu of possible formats from which to select, each format only encodes and decodes a single resolution and frame rate. Because the display or motion rates of these formats are not integrally related to each other, conversion from one to another is difficult.
Further, this proposal does not provide a crucial capability of compatibility with computer displays. These proposed image motion rates are based upon historical rates which date back to the early part of this century. If a "clean-slate" were to be made, it is unlikely that these rates would be chosen. In the computer industry, where displays could utilize any rate over the last decade, rates in the 70 to 80 Hz range have proven optimal, with 72 and 75 Hz being the most common rates. Unfortunately, the proposed rates of 30 5 and 60 Hz lack useful interoperability with 72 or 75 Hz, resulting in degraded temporal performance.
In addition, it is being suggested by some that interlace is required, due to a claimed need to have about 1000 lines of resolution at high frame rates, but based upon the notion that such images cannot be compressed within the available 18-19 o mbits/second of a conventional 6 MHz broadcast television channel.
It would be much more desirable if a single signal format were to be adopted, containing within it all ofthe desired standard and high definition resolutions. However, to do so within the bandwidth constraints of a conventional 6 MHz broadcast television channel requires compression and "scalability" of both frame rate (temporal) and 5 resolution (spatial). One method specifically intended to provide for such scalability is the MPEG-2 standard. Unfortunately, the temporal and spatial scalability features specified within the MPEG-2 standard (and newer standards, like MPEG-4) are not sufficiently efficient to accommodate the needs of advanced television for the U.S. Thus, the proposal for advanced television for the U.S . is based upon the premise that temporal 0 (frame rate) and spatial (resolution) layering are inefficient, and therefore discrete formats are necessary.
Further, it would be desirable to provide enhancements to resolution, image clarity, coding efficiency, and video production efficiency. The present invention provides such enhancements.
SUMMARY
The invention provides a method and apparatus for image compression which demonstrably achieves better than 1000-line resolution image compression at high frame rates with high quality. It also achieves both temporal and resolution scalability at this resolution at high frame rates within the available bandwidth of a conventional television broadcast channel. The inventive technique efficiently achieves over twice the compression ratio being proposed for advanced television. Further, layered compression allows a form of modularized decomposition of an image that supports flexible application of a variety of image enhancement techniques.
Image material is preferably captured at an initial or primary framing rate of 72 fps. An MPEG-like (e.g., MPEG-2, MPEG-4, etc.) data stream is then generated, comprising:
(1) a base layer, preferably encoded using only MPEG-type P frames, comprising a low resolution (e.g., 1024x512 pixels), low frame rate (24 or 36 Hz) bitstream;
(2) an optional base resolution temporal enhancement layer, encoded using only MPEG-type B frames, comprising a low resolution (e.g., 1024x512 pixels), high frame rate (72 Hz) bitstream;
(3) an optional base temporal high resolution enhancement layer, preferably encoded using only MPEG-type P frames, comprising a high resolution (e.g., 2kxlk pixels), low frame rate (24 or 36 Hz) bitstream;
(4) an optional high resolution temporal enhancement layer, encoded using only MPEG-type B frames, comprising a high resolution (e.g., 2kxlk pixels), high frame rate (72 Hz) bitstream.
The invention provides a number of key technical attributes, allowing substantial improvement over current proposals, and including: replacement of numerous resolutions and frame rates with a single layered resolution and frame rate; no need for interlace in order to achieve better than 1000-lines of resolution for 2 megapixel images at high frame rates (72 Hz) within a 6 MHz television channel; compatibility with computer displays through use of a primary framing rate of 72 fps; and greater robustness than the current unlayered format proposal for advanced television, since all available bits may be allocated to a lower resolution base layer when "stressful" image material is encountered.
Further, the invention provides a number of enhancements to handle a variety of video quality and compression problems. The following describes a number of such enhancements, most of which are preferably embodied as a set of tools which can be applied to the tasks of enhancing images and compressing such images. The tools can be combined by a content developer in various ways, as desired, to optimize the visual quality and compression efficiency of a compressed data stream, particularly a layered compressed data stream. Such tools include improved image filtering techniques, motion vector representation and determination, de-interlacing and noise reduction enhancements, motion analysis, imaging device characterization and correction, an enhanced 3-2 pulldown system, frame rate methods for production, a modular bit rate technique, a multi-layer DCT structure, variable length coding optimization, an augmentation system for MPEG-2 and MPEG-4, and guide vectors for the spatial enhancement layer.
The details of one or more embodiments ofthe invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages ofthe invention will be apparent from the description and drawings, and from the claims.
DESCRIPTION OF DRAWINGS
FIG. 1 is a timing diagram showing the pulldown rates for 24 fps and 36 fps material to be displayed at 60 Hz. FIG. 2 is a first preferred MPEG-2 coding pattern.
FIG. 3 is a second preferred MPEG-2 coding pattern.
FIG.4 is a block diagram showing temporal layer decoding in accordance with the preferred embodiment ofthe invention.
FIG. 5 is a block diagram showing 60 Hz interlaced input to a converter that can output both 36 Hz and 72 Hz frames.
FIG. 6 is a diagram showing a "master template" for a base MPEG-2 layer at 24 or 36 Hz.
FIG. 7 is a diagram showing enhancement of a base resolution template using hierarchical resolution scalability utilizing MPEG-2. FIG. 8 is a diagram showing the preferred layered resolution encoding process.
FIG. 9 is a diagram showing the preferred layered resolution decoding process.
FIG. 10 is a block diagram showing a combination of resolution and temporal scalable options for a decoder in accordance with the invention.
FIG. 11 is diagram of a base layer expanded by using a gray area and enhancement to provide picture detail.
FIG. 12 is a diagram ofthe relative shape, amplitudes, and lobe polarity of a preferred downsizing filter.
FIGS. 13A and 13B are diagrams of the relative shape, amplitudes, and lobe polarity of a pair of preferred upsizing filters for upsizing by a factor of 2. FIG. 14A is a block diagram of an odd-field de-interlacer.
FIG. 14B is a block diagram of an even-field de-interlacer. FIG. 15 is a block diagram of a frame de-interlacer using three de-interlaced fields.
FIG. 16 is a block diagram of an additional layered mode based upon a 2/3 base layer. FIG. 17 is a diagram of one example of applying higher bit rates to modular portions of a compressed data stream.
FIG. 18 graphically illustrates the relationships of DCT harmonics between two resolution layers. FIG. 19 graphically illustrates the similar relationships of DCT harmonics between three resolution layers.
FIG. 20 is a diagram showing a set of matched DCT block sizes for multiple resolution layers.
FIG. 21 is a diagram showing examples of splitting of motion compensation macroblocks for determining independent motion vectors.
FIG. 22 is a block diagram showing an augmentation system for MPEG-2 type systems.
FIG. 23 is a diagram showing use of motion vectors from a base layer as guide vectors for a resolution enhancement layer, FIGS. 24A-24E are data flow diagrams showing on example professional level enhancement mode.
Like reference symbols in the various drawings indicate like elements.
DETAILED DESCRIPTION
Throughout this description, the preferred embodiment and examples shown should be considered as exemplars, rather than as limitations on the invention.
TEMPORAL AND RESOLUTION LAYERING
Goals Of A Temporal Rate Family
After considering the problems ofthe prior art, and in pursuing the invention, the following goals were defined for specifying the temporal characteristics of a future digital television system:
• Optimal presentation ofthe high resolution legacy of 24 frame-per-second films. • Smooth motion capture for rapidly moving image types, such as sports.
• Smooth motion presentation of sports and similar images on existing analog NTSC displays, as well as computer-compatible displays operating at 72 or 75 Hz.
• Reasonable but more efficient motion capture of less-rapidly-moving images, such as news and live drama. • Reasonable presentation of all new digital types of images through a converter box onto existing NTSC displays.
• High quality presentation of all new digital types of images on computer-compatible displays.
• If 60 Hz digital standard or high resolution displays come into the market, reasonable or high quality presentation on these displays as well.
Since 60 Hz and 72/75 Hz displays are fundamentally incompatible at any rate other than the movie rate of 24 Hz, the best situation would be if either 72/75 or 60 were eliminated as a display rate. Since 72 or 75 Hz is a required rate for N.I.I. (National Information Infrastructure) and computer applications, elimination ofthe 60 Hz rate as being fundamentally obsolete would be the most future-looking. However, there are many competing interests within the broadcasting and television equipment industries, and there is a strong demand that any new digital television infrastructure be based on 60 Hz (and 30 Hz). This has lead to much heated debate between the television, broadcast, and computer industries. Further, the insistence by some interests in the broadcast and television industries on interlaced 60 Hz formats further widens the gap with computer display requirements. Since non-interlaced display is required for computer-like applications of digital television systems, a de-interlacer is required when interlaced signals are displayed. There is substantial debate about the cost and quality of de-interlacers, since they would be needed in every such receiving device. Frame rate conversion, in addition to de-interlacing, further impacts cost and quality. For example, that NTSC to-from PAL converters continue to be very costly and yet conversion performance is not dependable for many common types of scenes. Since the issue of interlace is a complex and problematic subject, and in order to attempt to address the problems and issue of temporal rate, the invention is described in the context of a digital television standard without interlace.
Selecting Optimal Temporal Rates
Beat Problems. Optimal presentation on a 72 or 75 Hz display will occur if a camera or simulated image is created having a motion rate equal to the display rate (72 or 75 Hz, respectively), and vice versa. Similarly, optimal motion fidelity on a 60 Hz display will result from a 60 Hz camera or simulated image. Use of 72 Hz or 75 Hz generation rates with 60 Hz displays results in a 12 Hz or 15 Hz beat frequency, respectively. This beat can be removed through motion analysis, but motion analysis is expensive and inexact, often leading to visible artifacts and temporal aliasing. In the absence of motion analysis, the beat frequency dominates the perceived display rate, making the 12 or 15 Hz beat appear to provide less accurate motion than even 24 Hz. Thus, 24 Hz forms a natural temporal common denominator between 60 and 72 Hz. Although 75 Hz has a slightly higher 15 Hz beat with 60 Hz, its motion is still not as smooth as 24 Hz, and there is no integral relationship between 75 Hz and 24 Hz unless the 24 Hz rate is increased to 25 Hz. (In European 50 Hz countries, movies are often played 4% fast at 25 Hz; this can be done to make film presentable on 75 Hz displays.)
In the absence of motion analysis at each receiving device, 60 Hz motion on 72 or 75 Hz displays, and 75 or 72 Hz motion on 60 Hz displays, will be less smooth than 24 Hz images. Thus, neither 72/75 Hz nor 60 Hz motion is suitable for reaching a heterogeneous display population containing both 72 or 75 Hz and 60 Hz displays.
3-2 Pulldown. A further complication in selecting an optimal frame rate occurs due to the use of "3-2 pulldown" combined with video effects during the telecine (film- to-video) conversion process. During such conversions, the 3-2 pulldown pattern repeats a first frame (or field) 3 times, then the next frame 2 times, then the next frame 3 times, then the next frame 2 times, etc. This is how 24 fps film is presented on television at 60 Hz (actually, 59.94 Hz for NTSC color). That is, each of 12 pairs of 2 frames in one second of film is displayed 5 times, giving 60 images per second. The 3-2 pulldown pattern is shown in FIG. 1.
By some estimates, more than half of all film on video has substantial portions where adjustments have been made at the 59.94 Hz video field rate to the 24 fps film. Such adjustments include "pan-and-scan", color correction, and title scrolling. Further, many films are time-adjusted by dropping frames or clipping the starts and ends of scenes to fit within a given broadcast scheduled. These operations can make the 3-2 pulldown process impossible to reverse, since there is both 59.94 Hz and 24 Hz motion. This can make the film very difficult to compress using the MPEG-2 standard. Fortunately, this problem is limited to existing NTSC-resolution material, since there is no significant library of higher resolution digital film using 3-2 pulldown. Motion Blur, m order to further explore the issue of finding a common temporal rate higher than 24 Hz, it is useful to mention motion blur in the capture of moving images. Camera sensors and motion picture film are open to sensing a moving image for a portion of the duration of each frame. On motion picture cameras and many video cameras, the duration of this exposure is adjustable. Film cameras require a period of time tp advance the film, and are usually limited to being open only about 210 out of 360 degrees, or a 58% duty cycle. On video cameras having CCD sensors, some portion ofthe frame time is often required to "read" the image from the sensor. This can vary from 10% to 50% ofthe frame time. In some sensors, an electronic shutter must be used to blank the light during this readout time. Thus, the "duty cycle" of CCD sensors usually varies from 50 to 90%, and is adjustable in some cameras. The light shutter can sometimes be adjusted to further reduce the duty cycle, if desired. However, for both film and video, the most common sensor duty cycle duration is 50%.
Preferred Rate. With this issue in mind, one can consider the use of only some of the frames from an image sequence captured at 60, 72, or 75 Hz. Utilizing one frame in two, three, four, etc., the subrates shown in TABLE 1 can be derived.
Rate 1/2 Rate 1/3 Rate 1/4 Rate 1/5 Rate 1/6 Rate
75 Hz 37.5 25 18.25 15 12.5
72 Hz 36 24 18 14.4 12
60 Hz 30 20 15 12 10
TABLE 1
The rate of 15 Hz is a unifying rate between 60 and 75 Hz. The rate of 12 Hz is a unifying rate between 60 and 72 Hz. However, the desire for a rate above 24 Hz eliminates these rates.24 Hz is not common, but the use of 3-2 pulldown has come to be accepted by the industry for presentation on 60 Hz displays. The only candidate rates are therefore 30, 36, and 37.5 Hz. Since 30 Hz has a 7.5 Hz beat with 75 Hz, and a 6 Hz beat with 72 Hz, it is not useful as a candidate. The motion rates of 36 and 37.5 Hz become prime candidates for smoother motion than 24 Hz material when presented on 60 and 72/75 Hz displays. Both of these rates are about 50% faster and smoother than 24 Hz. The rate of 37.5 Hz is not suitable for use with either 60 or 72 Hz, so it must be eliminated, leaving only 36 Hz as having the desired temporal rate characteristics. (The motion rate of 37.5 Hz could be used if the 60 Hz display rate for television can be move 4% to 62.5 Hz. Given the interests behind 60 Hz, 62.5 Hz appears unlikely - there are even those who propose the very obsolete 59.94 Hz rate for new television systems. However, if such a change were to be made, the other aspects ofthe invention could be applied to the 37.5 Hz rate.)
The rates of 24, 36, 60, and 72 Hz are left as candidates for a temporal rate family. The rates of 72 and 60 Hz cannot be used for a distribution rate, since motion is less smooth when converting between these two rates than if 24 Hz is used as the distribution rate, as described above. By hypothesis, we are looking for a rate faster than 24 Hz. Therefore, 36 Hz is the prime candidate for a master, unifying motion capture and image distribution rate for use with 60 and 72/75 Hz displays.
As noted above, the 3-2 pulldown pattern for 24 Hz material repeats a first frame (or field) 3 times, then the next frame 2 times, then the next frame 3 times, then the next frame 2 times, etc. When using 36 Hz, each pattern optimally should be repeated in a 2-1-2 pattern. This can be seen in TABLE 2 and graphically in FIG. 1.
Rate Frame Numbers
60 Hz 1 2 3 4 5 6 7 8 9 10 24 Hz 1 1 1 2 2 3 3 3 4 4 36 Hz 1 1 2 3 3 4 4 5 6 6
TABLE 2
This relationship between 36 Hz and 60 Hz only holds for true 36 Hz material. 60 Hz material can be "stored" in 36 Hz, if it is interlaced, but 36 Hz cannot be reasonable created from 60 Hz without motion analysis and reconstruction. However, in looking for a new rate for motion capture, 36 Hz provides slightly smoother motion on 60 Hz than does 24 Hz, and provides substantially better image motion smoothness on a 72 Hz display. Therefore, 36 Hz is the optimum rate for a master, unifying motion capture and image distribution rate for use with 60 and 72 Hz displays, yielding smoother motion than 24 Hz material presented on such displays.
Although 36 Hz meets the goals set forth above, it is not the only suitable capture rate. Since 36 Hz cannot be simply extracted from 60 Hz, 60 Hz does not provide a suitable rate for capture. However, 72 Hz can be used for capture, with every other frame then used as the basis for 36 Hz distribution. The motion blur from using every other frame of 72 Hz material will be half of the motion blur at 36 Hz capture. Tests of motion blur appearance of every third frame from 72 Hz show that staccato strobing at 24 Hz is objectionable. However, utilizing every other frame from 72 Hz for 36 Hz display is not objectionable to the eye compared to 36 Hz native capture.
Thus, 36 Hz affords the opportunity to provide very smooth motion on 72 Hz displays by capturing at 72 Hz, while providing better motion on 60 Hz displays than 24 Hz material by using alternate frames of 72 Hz native capture material to achieve a 36 Hz distribution rate and then using 2-1-2 pulldown to derive a 60 Hz image. In summary, TABLE 3 shows the preferred optimal temporal rates for capture and distribution in accordance with the invention.
Preferred Rates
Capture Distribution Optimal Display Acceptable Display
72 Hz 36 Hz + 36 Hz 72 Hz 60 Hz
TABLE 3
It is also worth noting that this technique of utilizing alternate frames from a
72 Hz camera to achieve a 36 Hz distribution rate can profit from an increased motion blur duty cycle. The normal 50% duty cycle at 72 Hz, yielding a 25% duty cycle at 36 Hz, has been demonstrated to be acceptable, and represents a significant improvement over 24 Hz on 60 Hz and 72 Hz displays. However, if the duty cycle is increased to be in the 75-90% range, then the 36 Hz samples would begin to approach the more common 50% duty cycle. Increasing the duty rate may be accomplished, for example, by using "backing store" CCD designs which have a short blanking time, yielding a high duty cycle. Other methods may be used, including dual CCD multiplexed designs.
Modified MPEG-2 Compression For efficient storage and distribution, digital source material having the preferred temporal rate of 36 Hz should be compressed. The preferred form of compression for the invention is accomplished by using a novel variation ofthe MPEG-2 standard, but may be used with any other compression system with similar characteristics (e.g., MPEG-4).
MPEG-2 Basics. MPEG-2 is an international video compression standard defining a video syntax that provides an efficient way to represent image sequences in the form of more compact coded data. The language ofthe coded bits is the "syntax." For example, a few tokens can represent an entire block of 64 samples. MPEG also describes a decoding (reconstruction) process where the coded bits are mapped from the compact representation into the original, "raw" format ofthe image sequence. For example, a flag in the coded bitstream signals whether the following bits are to be decoded with a discrete cosine transform (DCT) algorithm or with a prediction algorithm. The algorithms comprising the decoding process are regulated by the semantics defined by MPEG. This syntax can be applied to exploit common video characteristics such as spatial redundancy, temporal redundancy, uniform motion, spatial masking, etc. In effect, MPEG-2 defines a programming language as well as a data format. An MPEG-2 decoder must be able to parse and decode an incoming data stream, but so long as the data stream complies with the MPEG-2 syntax, a wide variety of possible data structures and compression techniques can be used. The invention takes advantage of this flexibility by devising a novel means and method for temporal and resolution scaling using the MPEG-2 standard.
MPEG-2 uses an intraframe and an interframe method of compression. In most video scenes, the background* remains relatively stable while action takes place in the foreground. The background may move, but a great deal of the scene is redundant. MPEG-2 starts its compression by creating a reference frame called an I (for Intra) frame. I frames are compressed without reference to other frames and thus contain an entire frame of video information. I frames provide entry points into a data bitstream for random access, but can only be moderately compressed. Typically, the data representing I frames is placed in the bitstream every 10 to 15 frames. Thereafter, since only a small portion of the frames that fall between the reference I frames are different from the bracketing I frames, only the differences are captured, compressed and stored. Two type of frames are used for such differences - P (for Predicted) frames and B (for Bi-directional Interpolated) frames. P frames generally are encoded with reference to a past frame (either an I frame or a previous P frame), and, in general, will be used as a reference for future P frames. P frames receive a fairly high amount of compression. B frames pictures provide the highest amount of compression but generally require both a past and a future reference in order to be encoded. Bi-directional frames are never used for reference frames. Macroblocks within P frames may also be individually encoded using intra-frame coding. Macroblocks within B frames may also be individually encoded using intra-frame coding, forward predicted coding, backward predicted coding, or both forward and backward, or bi-directionally interpolated, predicted coding. A macroblock is a 16x16 pixel grouping of four 8x8 DCT blocks, together with one motion vector for P frames, and one or two motion vectors for B frames. After coding, an MPEG data bitstream comprises a sequence of I, P, and B frames. A sequence may consist of almost any pattern of I, P, and B frames (there are a few minor semantic restrictions on their placement). However, it is common in industrial practice to have a fixed pattern (e.g., IBBPBBPBBPBBPBB). As an important part of the invention, an MPEG-2 data stream is created comprising a base layer, at least one optional temporal enhancement layer, and an optional resolution enhancement layer. Each of these layers will be described in detail.
Temporal Scalability
Base Layer. The base layer is used to carry 36 Hz source material. In the preferred embodiment, one of two MPEG-2 frame sequences can be used for the base layer: IBPBPBP or IPPPPPP. The latter pattern is most preferred, since the decoder would only need to decode P frames, reducing the required memory bandwidth if 24 Hz movies were also decoded without B frames.
72 Hz Temporal Enhancement Layer. When using MPEG-2 compression, it is possible to embed a 36 Hz temporal enhancement layer as B frames within the MPEG-2 sequence for the 36 Hz base layer if the P frame distance is even. This allows the single data stream to support both 36 Hz display and 72 Hz display. For example, both layers could be decoded to generate a 72 Hz signal for computer monitors, while only the base layer might be decoded and converted to generate a 60 Hz signal for television. In the preferred embodiment, the MPEG-2 coding patterns of
IPBBBPBBBPBBBP or PBPBPBPB both allow placing alternate frames in a separate stream containing only temporal enhancement B frames to take 36 Hz to 72 Hz. These coding patterns are shown in FIG.S 2 and 3, respectively. The 2-Frame P spacing coding pattern of FIG. 3 has the added advantage that the 36 Hz decoder would only need to decode P frames, reducing the required memory bandwidth if 24 Hz movies were also decoded without B frames.
Experiments with high resolution images have suggested that the 2-Frame P spacing of FIG. 3 is optimal for most types of images. That is, the construction in FIG. 3 appears to offer the optimal temporal structure for supporting both 60 and 72 Hz, while providing excellent results on modern 72 Hz computer-compatible displays. This construction allows two digital streams, one at 36 Hz for the base layer, and one at 36 Hz for the enhancement layer B frames to achieve 72 Hz. This is illustrated in FIG.4, which is a block diagram showing that a 36 Hz base layer MPEG-2 decoder 50 simply decodes the P frames to generate 36 Hz output, which may then be readily converted to either 60 Hz or 72 Hz display. An optional second decoder 52 simply decodes the B frames to generate a second 36 Hz output, which when combined with the 36 Hz output ofthe base layer decoder 50 results in a 72 Hz output (a method for combining is discussed below). In an alternative embodiment, one fast MPEG-2 decoder 50 could decode both the P frames for the base layer and the B frames for the enhancement layer. Optimal Master Format. A number of companies are building MPEG-2 decoding chips which operate at around 11 MPixels/second. The MPEG-2 standard has defined some "profiles" for resolutions and frame rates. Although these profiles are strongly biased toward computer-incompatible format parameters such as 60 Hz, non-square pixels, and interlace, many chip manufacturers appear to be developing decoder chips which operate at the "main profile, main level". This profile is defined to be any horizontal resolution up to 720 pixels, any vertical resolution up to 576 lines at up to 25 Hz, and any frame rate of up to 480 lines at up to 30 Hz. A wide range of data rates from approximately 1.5 Mbits/second to about 10 Mbits/second is also specified. However, from a chip point of view, the main issue is the rate at which pixels are decoded. The main-level, main-profile pixel rate is about 10.5 MPixels/second.
Although there is variation among chip manufacturers, most MPEG-2 decoder chips will in fact operate at up to 13 MPixels/second, given fast support memory. Some decoder chips will go as fast as 20 MPixels/second or more. Given that CPU chips tend to gain 50% improvement or more each year at a given cost, one can expect some near-term flexibility in the pixel rate of MPEG-2 decoder chips.
TABLE 4 illustrates some desirable resolutions and frame rates, and their corresponding pixel rates.
Resolution Frame Rate Pixel Rate
X Y (Hz) (MPixels/s)
640 480 36 11.1
720 486 36 12.6
720 486 30 (for comparison) 10.5 Resolution Frame Rate Pixel Rate
X Y (Hz) (MPixels/s)
704 480 36 12.2
704 480 30 (for comparison) 10.1
680 512 36 12.5
1024 512 24 12.6
TABLE 4
All of these formats can be utilized with MPEG-2 decoder chips that can generate at least 12.6 MPixels/second. The very desirable 640x480 at 36 Hz format can be achieved by nearly all current chips, since its rate is 11.1 MPixels/second. A widescreen 1024x512 image can be squeezed into 680x512 using a 1.5:1 squeeze, and can be supported at 36 Hz if 12.5 MPixels/second can be handled. The highly desirable square pixel widescreen template of 1024x512 can achieve 36 Hz when MPEG-2 decoder chips can process about 18.9 MPixels/second. This becomes more feasible if 24 Hz and 36 Hz material is coded only with P frames, such that B frames are only required in the 72 Hz temporal enhancement layer decoders. Decoders which use only P frames require less memory and memory bandwidth, making the goal of 19 MPixels/second more accessible.
The 1024x512 resolution template would most often be used with 2.35:1 and
1.85:1 aspect ratio films at 24 fps. This material only requires 11.8 MPixels/second, which should fit within the limits of most existing main level-main profile decoders. All of these formats are shown in FIG.6 in a "master template" for a base layer at
24 or 36 Hz. Accordingly, the invention provides a unique way of accommodating a wide variety of aspect ratios and temporal resolution compared to the prior art. (Further discussion of a master template is set forth below).
The temporal enhancement layer of B frames to generate 72 Hz can be decoded using a chip with double the pixel rates specified above, or by using a second chip in parallel with additional access to the decoder memory. Under the invention, at least two ways exist for merging of the enhancement and base layer data streams to insert the alternate B frames. First, merging can be done invisibly to the decoder chip using the MPEG-2 transport layer. The MPEG-2 transport packets for two PIDs (Program IDs) can be recognized as containing the base layer and enhancement layer, and their stream contents can both be simply passed on to a double-rate capable decoder chip, or to an appropriately configured pair of normal rate decoders. Second, it is also possible to use the "data partitioning" feature in the MPEG-2 data stream instead ofthe transport layer from MPEG-2 systems. The data partitioning feature allows the B frames to be marked as belonging to a different class within the MPEG-2 compressed data stream, and can therefore be flagged to be ignored by 36-Hz decoders which only support the temporal base layer rate.
Temporal scalability, as defined by MPEG-2 video compression, is not as optimal as the simple B frame partitioning ofthe invention. The MPEG-2 temporal scalability is only forward referenced from a previous P or B frame, and thus lacks the efficiency available in the B frame encoding proposed here, which is both forward and backward referenced. Accordingly, the simple use of B frames as a temporal enhancement layer provides a simpler and more efficient temporal scalability than does the temporal scalability defined within MPEG-2. Notwithstanding, this use of B frames as the mechanism for temporal scalability is fully compliant with MPEG-2. The two methods of identifying these B frames as an enhancement layer, via data partitioning or alternate PID's for the B frames, are also fully compliant.
50/60 Hz Temporal enhancement layer. In addition to, or as an alternative to, the 72 Hz temporal enhancement layer described above (which encodes a 36 Hz signal), a 60 Hz temporal enhancement layer (which encodes a 24 Hz signal) can be added in similar fashion to the 36 Hz base layer. A 60 Hz temporal enhancement layer is particular useful for encoding existing 60 Hz interlaced video material.
Most existing 60 Hz interlaced material is video tape for NTSC in analog, Dl, or D2 format. There is also a small amount of Japanese HDTV (SMPTE 240/260M). There are also cameras which operate in this format Any such 60 Hz interlaced format can be processed in known fashion such that the signal is de-interlaced and frame rate converted. This process involves very complex image understanding technology, similar to robot vision. Even with very sophisticated technology, temporal aliasing generally will result in "misunderstandings" by the algorithm and occasionally yield artifacts. Note that the typical 50% duty cycle of image capture means that the camera is "not looking" half the time. The "backwards wagon wheels" in movies is an example of temporal aliasing due to this normal practice of temporal undersampling. Such artifacts generally cannot be removed without human-assisted reconstruction. Thus, there will always be cases which cannot be automatically corrected. However, the motion conversion results available in current technology should be reasonable on most material.
The price of a single high definition camera or tape machine would be similar to the cost of such a converter. Thus, in a studio having several cameras and tape machines, the cost of such conversion becomes modest. However, performing such processing adequately is presently beyond the budget of home and office products. Thus, the complex processing to remove interlace and convert the frame rate for existing material is preferably accomplished at the origination studio. This is shown in FIG. 5, which is a block diagram showing 60 Hz interlaced input from cameras 60 or other sources (such as non-film video tape) 62 to a converter 64 that includes a de-interlacer function and a frame rate conversion function that can output a 36 Hz signal (36 Hz base layer only) and a 72 Hz signal (36 Hz base layer plus 36 Hz from the temporal enhancement layer).
As an alternative to outputting a 72 Hz signal (36 Hz base layer plus 36 Hz from the temporal enhancement layer), this conversion process can be adapted to produce a second MPEG-224 Hz temporal enhancement layer on the 36 Hz base layer which would reproduce the original 60 Hz signal, although de-interlaced. If similar quantization is used for the 60 Hz temporal enhancement layer B frames, the data rate should be slightly less than the 72 Hz temporal enhancement layer, since there are fewer B frames. >60 I → 36 + 36 = 72
>601 → 36 + 24 = 60
>72 → 36, 72, 60
>50 I → 36, 50, 72
>60 → 24, 36, 72
The vast majority of material of interest to the United States is low resolution
NTSC. At present, most NTSC signals are viewed with substantial impairment on most home televisions. Further, viewers have come to accept the temporal impairments inherent in the use of 3-2 pulldown to present film on television. Nearly all prime-time television is made on film at 24 frames per second. Thus, only sports, news, and other video-original shows need be processed in this fashion. The artifacts and losses associated with converting these shows to a 36/72 Hz format are likely to be offset by the improvements associated with high-quality de-interlacing ofthe signal.
Note that the motion blur inherent in the 60 Hz (or 59.94 Hz) fields should be very similar to the motion blur in 72 Hz frames. Thus, this technique of providing abase
5 and enhancement layer should appear similar to 72 Hz origination in terms of motion blur. Accordingly, few viewers will notice the difference, except possibly as a slight improvement, when interlaced 60 Hz NTSC material is processed into a 36 Hz base layer, plus 24 Hz from the temporal enhancement layer, and displayed at 60 Hz. However, those who buy new 72 Hz digital non-interlaced televisions will notice a small improvement o when viewing NTSC, and a major improvement when viewing new material captured or originated at 72 Hz. Even the decoded 36 Hz base layer presented on 72 Hz displays will look as good as high quality digital NTSC, replacing interlace artifacts with a slower frame rate.
The same process can also be applied to the conversion of existing PAL 50 Hz 5 material to a second MPEG-2 enhancement layer. PAL video tapes are best slowed to 48 Hz prior to such conversion. Live PAL requires conversion using the relatively unrelated rates of 50, 36, and 72 Hz. Such converter units presently are only affordable at the source of broadcast signals, and are not presently practical at each receiving device in the home and office.
0 Resolution Scalability
It is possible to enhance the base resolution template using hierarchical resolution scalability utilizing MPEG-2 to achieve higher resolutions built upon a base layer. Use of enhancement can achieve resolutions at 1.5x and 2x the base layer. Double resolution can be built in two steps, by using 3/2 then 4/3, or it can be a single factor-of-two step. This is 5 shown in FIG. 7.
The process of resolution enhancement can be achieved by generating a resolution enhancement layer as an independent MPEG-2 stream and applying MPEG-2 compression to the enhancement layer. This technique differs from the "spatial scalability" defined with MPEG-2, which has proven to be highly inefficient. However, MPEG-2 0 contains all ofthe tools to construct an effective layered resolution to provide spatial scalability. The preferred layered resolution encoding process ofthe invention is shown in FIG. 8. The preferred decoding process ofthe invention is shown in FIG. 9.
Resolution Layer Coding. In FIG. 8, an original 2kxlk image 80 is down-filtered, preferably using an optimized filter having negative lobes (see discussion of FIG. 12 below), to 112 resolution in each dimension to create a 1024x512 base layer 81. The base layer 81 is then compressed according to conventional MPEG-2 algorithms, generating an MPEG-2 base layer 82 suitable for transmission. Importantly, full MPEG-2 motion compensation can be used during this compression step. That same signal is then decompressed using conventional MPEG-2 algorithms back to a 1024x512 image 83. The 1024x512 image 83 is expanded (for example, by pixel replication, or preferably by better up-filters such as spline interpolation or filters having negative lobes; see discussion of FIGS. 13A and 13B below) to a first 2kxlk enlargement 84.
Meanwhile, as an optional step, the filtered 1024x512 base layer 81 is expanded to a second 2kxlk enlargement 85. This second 2kxlk enlargement 85 is subtracted from the original 2kxlk image 80 to generate an image that represents the top octave of resolution between the original high resolution image 80 and the original base layer image 81. The resulting image is optionally multiplied by a sharpness factor or weight, and then added to the difference between the original 2kxlk image 80 and the second 2kxlk enlargement 85 to generate a center-weighted 2kxlk enhancement layer source image 86. This enhancement layer source image 86 is then compressed according to conventional MPEG-2 algorithms, generating a separate MPEG-2 resolution enhancement layer 87 suitable for transmission. Importantly, full MPEG-2 motion compensation can be used during this compression step.
Resolution Layer Decoding. In FIG. 9, the base layer 82 is decompressed using conventional MPEG-2 algorithms back to a 1024x512 image 90. The 1024x512 image 90 is expanded to a first 2kxlk image 91. Meanwhile, the resolution enhancement layer 87 is decompressed using conventional MPEG-2 algorithms back to a second 2kxlk image 92. The first 2kxlk image 91 and the second 2kxlk image 92 are then added to generate a high-resolution 2kxlk image 93. Improvements Over MPEG-2. In essence, the enhancement layer is created by expanding the decoded base layer, taking the difference between the original image and the decode base layer, and compressing. However, a compressed resolution enhancement layer may be optionally added to the base layer after decoding to create a higher resolution image in the decoder. The inventive layered resolution encoding process differs from MPEG-2 spatial scalability in several ways: • The enhancement layer difference picture is compressed as its own MPEG-2 data stream, with I, B, and P frames. This difference represents the major reason that resolution scalability, as proposed here, is effective, where MPEG-2 spatial scalability is ineffective. The spatial scalability defined within MPEG-2 allows an upper layer to be coded as the difference between the upper layer picture and the expanded base layer, or as a motion compensated MPEG-2 data stream ofthe actual picture, or a combination of both. However, neither of these encodings is efficient.
The difference from the base layer could be considered as an I frame of the difference, which is inefficient compared to a motion-compensated difference picture, as in the invention. The upper-layer encoding defined within MPEG-2 is also inefficient, since it is identical to a complete encoding ofthe upper layer. The motion compensated encoding of the difference picture, as in the invention, is therefore substantially more efficient.
• Since the enhancement layer is an independent MPEG-2 data stream, the MPEG-2 systems transport layer (or another similar mechanism) must be used to multiplex the base layer and enhancement layer.
• The expansion and resolution reduction (down) filtering can be a gaussian or spline function, or a filter with negative lobes (see FIG. 12), which are more optimal than the bilinear interpolation specified in MPEG-2 spatial scalability.
• The image aspect ratio must match between the lower and higher layers in the preferred embodiment. In MPEG-2 spatial scalability, extensions to width and/or height are allowed. Such extensions are not allowed in the preferred embodiment due to efficiency requirements.
• Due to efficiency requirements, and the extreme amounts of compression used in the enhancement layer, the entire area ofthe enhancement layer is not coded. Usually, the area excluded from enhancement will be the border area. Thus, the 2kxlk enhancement layer source image 86 in the preferred embodiment is center-weighted. In the preferred embodiment, a fading function (such as linear weighting) is used to "feather" the enhancement layer toward the center ofthe image and away from the border edge to avoid abrupt transitions in the image. Moreover, any manual or automatic method of determining regions having detail which the eye will follow can be utilized to select regions which need detail, and to exclude regions where extra detail is not required. All ofthe image has detail to the level ofthe base layer, so all ofthe image is present. Only the areas of special interest benefit from the enhancement layer. In the absence of other criteria, the edges or borders ofthe frame can be excluded from enhancement, as in the center-weighted embodiment described above. The MPEG-2 parameters "lower_layer_prediction_horizontal&vertical offset" parameters used as signed negative integers, combined with the "horizontal&ver- tical_subsampling_factor_m&n" values, can be used to specify the enhancement layer rectangle's overall size and placement within the expanded base layer.
• A sharpness factor is added to the enhancement layer to offset the loss of sharpness which occurs during quantization. Care must be taken to utilize this parameter only to restore the clarity and sharpness of the original picture, and not to enhance the image. As noted above with respect to FIG. 8, the sharpness factor is the "high octave" of resolution between the original high resolution image 80 and the original base layer image 81 (after expansion). This high octave image will be quite noisy, in addition to containing the sharpness and detail of the high octave of resolution.
Adding too much of this image can yield instability in the motion compensated encoding ofthe enhancement layer. The amount that should be added depends upon the level ofthe noise in the original image. A typical weighting value is 0.25. For noisy images, no sharpness should be added, and it even may be advisable to suppress the noise in the original for the enhancement layer before compressing using conventional noise suppression techniques which preserve detail.
• Temporal and resolution scalability are intermixed by utilizing B frames for temporal enhancement from 36 to 72 Hz in both the base and resolution enhancement layers. In this way, four possible levels of decoding performance are possible with two layers of resolution scalability, due to the options available with two levels of temporal scalability. These differences represent substantial improvements over MPEG-2 spatial and temporal scalability. However, these differences are still consistent with MPEG-2 decoder chips, although additional logic may be required in the decoder to perform the expansion and addition in the resolution enhancement decoding process shown in FIG. 9. Such
5 additional logic is nearly identical to that required by the less effective MPEG-2 spatial scalability.
Optional Non-MPEG-2 Coding of the Resolution Enhancement Layer. It is possible to utilize a different compression technique for the resolution enhancement layer than MPEG-2. Further, it is not necessary to utilize the same compression technology for o the resolution enliancement layer as for the base layer. For example, motion-compensated block wavelets can be utilized to match and track details with great efficiency when the difference layer is coded. Even if the most efficient position for placement of wavelets jumps around on the screen due to changing amounts of differences, it would not be noticed in the low-amplitude enhancement layer. Further, it is not necessary to cover the 5 entire image - it is only necessary to place the wavelets on details. The wavelets can have their placement guided by detail regions in the image. The placement can also be biased away from the edge.
Multiple Resolution Enhancement Layers. At the bit rates being described here, where 2 MPixels (2048x1024) at 72 frames per second are being coded in 18.5 0 mbits/second, only a base layer ( 1024x512 at 72fps) and a single resolution enhancement layer have been successfully demonstrated. However, the anticipated improved efficiencies available from further refinement of resolution enhancement layer coding should allow for multiple resolution enhancement layers. For example, it is conceivable that a base layer at 512x256 could be resolution-enhanced by four layers to 1024x512, 5 1536x768, and 2048x1024. This is possible with existing MPEG-2 coding at the movie frame rate of 24 frames per second. At high frame rates such as 72 frames per second, MPEG-2 does not provide sufficient efficiency in the coding of resolution-enhancement layers to allow this many layers at present. Mastering Formats
Utilizing a template at or near 2048x1024 pixels, it is possible to create a single digital moving image master format source for a variety of release formats. As shown in
FIG.6, a 2k Ik template can efficiently support the common widescreen aspect ratios of 1.85:1 and 2.35:1. A 2kxlk template can also accommodate 1.33:1 and other aspect ratios.
Although integers (especially the factor of 2) and simple fractions (3/2 & 4/3) are most efficient step sizes in resolution layering, it is also possible to use arbitrary ratios to achieve any required resolution layering. However, using a 2048x1024 template, or something near it, provides not only a high quality digital master format, but also can provide many other convenient resolutions from a factor of two base layer (lkx512), including NTSC, the U.S. television standard.
It is also possible to scan film at higher resolutions such as 4kx2k, 4kx3k, or 4kx4k. Using optional resolution enhancement, these higher resolutions can be created from a central master format resolution near 2kxlk. Such enhancement layers for film will consist of both image detail, grain, and other sources of noise (such as scanner noise). Because of this noisiness, the use of compression technology in the enhancement layer for these very high resolutions will require alternatives to MPEG-2 types of compression. Fortunately, other compression technologies exist which can be utilized for compressing such noisy signals, while still maintaining the desired detail in the image. One example of such a compression technology is motion compensated wavelets or motion compensated fractals.
Preferably, digital mastering formats should be created in the frame rate ofthe film if from existing movies (i. e. , at 24 frames per second) . The common use of both 3 -2 pulldown and interlace would be inappropriate for digital film masters. For new digital electronic material, it is hoped that the use of 60 Hz interlace will cease in the near future, and be replaced by frame rates which are more compatible with computers, such as 72 Hz, as proposed here. The digital image masters should be made at whatever frame rate the images are captured, whether at 72 Hz, 60 Hz, 36 Hz, 37.5 Hz, 75 Hz, 50 Hz, or other rates. The concept of a mastering format as a single digital source picture format for all electronic release formats differs from existing practices, where PAL, NTSC, letterbox, pan-and-scan, HDTV, and other masters are all generally independently made from a film original. The use of a mastering format allows both film and digital/electronic shows to be mastered once, for release on a variety of resolutions and formats.
Combined Resolution and Temporal Enhancement Layers
As noted above, both temporal and resolution enhancement layering can be combined. Temporal enhancement is provided by decoding B frames. The resolution enhancement layer also has two temporal layers, and thus also contains B frames. For 24 fps film, the most efficient and lowest cost decoders might use only
P frames, thereby minimizing both memory and memory bandwidth, as well as simplifying the decoder by eliminating B frame decoding. Thus, in accordance with the invention, decoding movies at 24 fps and decoding advanced television at 36 fps could utilize a decoder without B frame capability. B frames can then be utilized between P frames to yield the higher temporal layer at 72 Hz, as shown in FIG. 3, which could be decoded by a second decoder. This second decoder could also be simplified, since it would only have to decode B frames.
Such layering also applies to the enhanced resolution layer, which can similarly utilize only P and I frames for 24 and 36 fps rates. The resolution enhancement layer can add the full temporal rate of 72 Hz at high resolution by adding B frame decoding within the resolution enhancement layer.
The combined resolution and temporal scalable options for a decoder are illustrated in FIG. 10. This example also shows an allocation ofthe proportions of an approximately 18 mbits/second data stream to achieve the spatio-temporal layered Advanced Television ofthe invention.
In FIG. 10, a base layer MPEG-2 1024x512 pixel data stream (comprising only P frames in the preferred embodiment) is applied to a base resolution decoder 100. Approximately 5 mbits/per sec of bandwidth is required for the P frames. The base resolution decoder 100 can decode at 24 or 36 fps. The output of the base resolution decoder 100 comprises low resolution, low frame rate images (1024x512 pixels at 24 or 36 Hz).
The B frames from the same data stream are parsed out and applied to a base resolution temporal enhancement layer decoder 102. Approximately 3 mbits/per sec of bandwidth is required for such B frames. The output ofthe base resolution decoder 100 is also coupled to the temporal enhancement layer decoder 102. The temporal enhancement layer decoder 102 can decode at 36 fps. The combined output of the temporal enhancement layer decoder 102 comprises low resolution, high frame rate images (1024x512 pixels at 72 Hz). Also in FIG. 10, a resolution enhancement layer MPEG-22kx 1 k pixel data stream
(comprising only P frames in the preferred embodiment) is applied to a base temporal high resolution enhancement layer decoder 104. Approximately 6 mbits/per sec of bandwidth is required for the P frames. The output ofthe base resolution decoder 100 is also coupled to the high resolution enhancement layer decoder 104. The high resolution enhancement layer decoder 104 can decode at 24 or 36 fps. The output of the high resolution enhancement layer decoder 104 comprises high resolution, low frame rate images (2kxlk pixels at 24 or 36 Hz).
The B frames from the same data stream are parsed out and applied to a high resolution temporal enhancement layer decoder 106. Approximately 4 mbits/per sec of bandwidth is required for such B frames. The output ofthe high resolution enhancement layer decoder 104 is coupled to the high resolution temporal enhancement layer decoder 106. The output ofthe temporal enhancement layer decoder 102 is also coupled to the high resolution temporal enhancement layer decoder 106. The high resolution temporal enhancement layer decoder 106 can decode at 36 fps. The combined output ofthe high resolution temporal enhancement layer decoder 106 comprises high resolution, high frame rate images (2kxlk pixels at 72 Hz).
Note that the compression ratio achieved through this scalable encoding mechanism is very high, indicating excellent compression efficiency. These ratios are shown in TABLE 5 for each ofthe temporal and scalability options from the example in FIG. 10. These ratios are based upon source RGB pixels at 24 bits/pixel. (If the 16 bits/pixel of conventional 4:2:2 encoding or the 12 bits/pixel of conventional 4:2:0 encoding are factored in, then the compression ratios would be 3/4 and 1/2, respectively, ofthe values shown.)
Layer Resolution Rate Data Rate - mb/s MPixels/s Comp.
(Hz) (typical) Ratio
(typical)
Base Ikx512 36 5 18.9 90
Base Temp. lkx512 72 8 (5+3) 37.7 113
High 2kxlk 36 11 (5+6) 75.5 165
High Temp. 2kxlk 72 18 (5+3+6+4) 151 201 for comparison:
CCI 601 720x486 29.97 5 10.5 50
TABLE 5
These high compression ratios are enabled by two factors:
1) The high temporal coherence of high-frame-rate 72 Hz images;
2) The high spatial coherence of high resolution 2kxlk images;
3) Application of resolution detail enhancement to the important parts ofthe image (e.g., the central heart), and not to the less important parts (e.g., the borders ofthe frame).
These factors are exploited in the inventive layered compression technique by taking advantage of the strengths of the MPEG-2 encoding syntax. These strengths include bi-directionally interpolated B frames for temporal scalability. The MPEG-2 syntax also provides efficient motion representation through the use of motion-vectors in both the base and enhancement layers. Up to some threshold of high noise and rapid image change, MPEG-2 is also efficient at coding details instead of noise within an enhancement layer through motion compensation in conjunction with DCT quantization. Above this threshold, the data bandwidth is best allocated to the base layer. These MPEG-2 mechanisms work together when used according to the invention to yield highly efficient and effective coding which is both temporally and spatially scalable.
In comparison to 5 mbits/second encoding of CCIR 601 digital video, the compression ratios in TABLE 5 are much higher. One reason for this is the loss of some coherence due to interlace. Interlace negatively affects both the ability to predict subsequent frames and fields, as well as the correlation between vertically adjacent pixels. Thus, a major portion ofthe gain in compression efficiency described here is due to the absence of interlace.
The large compression ratios achieved by the invention can be considered from the perspective ofthe number of bits available to code each MPEG-2 macroblock. As noted above, macroblock is a 16x16 pixel grouping of four 8x8 DCT blocks, together with one motion vector for P frames, and one or two motion vectors for B frames. The bits available per macroblock for each layer are shown in TABLE 6.
Layer Data Rate - mb/s MPixels/s Average Available
(typical) Bits Macroblock
Base 5 19 68
Base Temporal 8 (5+3) 38 54
High 11 (5+6) 76 37 overall, 20/enh. layer
High w/border 11 (5+6) 61 46 overall, 35/enh. layer around hi-res center
High Temporal 18 (5+3+6+4) 151 30 overall, 17/enh. layer
High Temporal 18 (5+3+6+4) 123 37 overall, 30/enh. layer w/border around hi-res center for comparison:
CCIR 601 5 10.5 122 TABLE 6
The available number of bits to code each macroblock is smaller in the enhancement layer than in the base layer. This is appropriate, since it is desirable for the base layer to have as much quality as possible. The motion vector requires 8 bits or so, leaving 10 to 25 bits for the macroblock type codes and for the DC and AC coefficients for all four 8x8 DCT blocks. This leaves room for only a few "strategic" AC coefficients. Thus, statistically, most of the information available for each macroblock must come from the previous frame of an enhancement layer.
It is easily seen why the MPEG-2 spatial scalability is ineffective at these compression ratios, since there is not sufficient data space available to code enough DC and AC coefficients to represent the high octave of detail represented by the enhancement difference image. The high octave is represented primarily in the fifth through eighth horizontal and vertical AC coefficients. These coefficients cannot be reached if there are only a few bits available per DCT block.
The system described here gains its efficiency by utilizing motion compensated prediction from the previous enhancement difference frame. This is demonstrably effective in providing excellent results in temporal and resolution (spatial) layered encoding.
Graceful Degradation The temporal scaling and resolution scaling techniques described here work well for normal-running material at 72 frames per second using a 2k lk original source. These techniques also work well on film-based material which runs at 24 φs. At high frame rates, however, when a very noise-like image is coded, or when there are numerous shot cuts within an image stream, the enhancement layers may lose the coherence between frames which is necessary for effective coding. Such loss is easily detected, since the buffer-fullness/rate-control mechanism of a typical MPEG-2 encoder/decoder will attempt to set the quantizer to very coarse settings. When this condition is encountered, all of the bits normally used to encode the resolution enhancement layers can be allocated to the base layer, since the base layer will need as many bits as possible in order to code the stressful material. For example, at between about 0.5 and 0.33 MPixels per frame for the base layer, at 72 frames per second, the resultant pixel rate will be 24 to 36 MPixels/second. Applying all ofthe available bits to the base layer provides about 0.5 to 0.67 million additional bits per frame at 18.5 mbits/second, which should be sufficient to code very well, even on stressful material.
Under more extreme cases, where every frame is very noise-like and/or there are cuts happening every few frames, it is possible to gracefully degrade even further without loss of resolution in the base layer. This can be done by removing the B frames coding the temporal enhancement layer, and thus allow use of all ofthe available bandwidth (bits) for the I and P frames of the base layer at 36 φs. This increases the amount of data available for each base layer frame to between about 1.0 and 1.5 mbits/frame (depending on the resolution ofthe base layer). This will still yield the fairly good motion rendition rate of 36 φs at the fairly high quality resolution ofthe base layer, under what would be extremely stressful coding conditions. However, if the base-layer quantizer is still operating at a coarse level under about 18.5 mbits/second at 36 φs, then the base layer frame rate can be dynamically reduced to 24, 18, or even 12 frames per second (which would make available between 1.5 and 4 mbits for every frame), which should be able to handle even the most pathological moving image types. Methods for changing frame rate in such circumstances are known in the art.
The current proposal for U.S. advanced television does not allow for these methods of graceful degradation, and therefore cannot perform as well on stressful material as the inventive system.
In most MPEG-2 encoders, the adaptive quantization level is controlled by the output buffer fullness. At the high compression ratios involved in the resolution enhancement layer ofthe invention, this mechanism may not function optimally. Various techniques can be used to optimize the allocation of data to the most appropriate image regions. The conceptually simplest technique is to perform a pre-pass of encoding over the resolution enhancement layer to gather statistics and to search out details which should be preserved. The results from the pre-pass can be used to set the adaptive quantization to optimize the preservation of detail in the resolution enhancement layer. The settings can also be artificially biased to be non-uniform over the image, such that image detail is biased to allocation in the main screen regions, and away from the macroblocks at the extreme edges ofthe frame. Except for leaving an enhancement-layer border at high frame rates, none of these adjustments are required, since existing decoders function well without such improvements. However, these further improvements are available with a small extra effort in the enhancement layer encoder.
Conclusion The choice of 36 Hz as a new common ground temporal rate appears to be optimal. Demonstrations ofthe use of this frame rate indicate that it provides significant improvement over 24 Hz for both 60 Hz and 72 Hz displays. Images at 36 Hz can be created by utilizing every other frame from 72 Hz image capture. This allows combining a base layer at 36 Hz (preferably using P frames) and a temporal enhancement layer at 36 Hz (using B frames) to achieve a 72 Hz display. The "future-looking" rate of 72 Hz is not compromised by the inventive approach, while providing transition for 60 Hz analog NTSC display. The invention also allows a transition for other 60 Hz displays, if other passive-entertainment-only (computer incompatible) 60 Hz formats under consideration are accepted. Resolution scalability can be achieved though using a separate MPEG-2 image data stream for a resolution enhancement layer. Resolution scalability can take advantage ofthe B frame approach to provide temporal scalability in both the base resolution and enhancement resolution layers.
The invention described here achieves many highly desirable features. It has been claimed by some involved in the U.S . advanced television process that neither resolution nor temporal scalability can be achieved at high definition resolutions within the approximately 18.5 mbits/second available in terrestrial broadcast. However, the invention achieves both temporal and spatial-resolution scalability within this available data rate. It has also been claimed that 2 MPixels at high frame rates cannot be achieved without the use of interlace within the available 18.5 mbits/second data rate. However, achieves not only resolution (spatial) and temporal scalability, it can provide 2 MPixels at 72 frames per second.
In addition to providing these capabilities, the invention is also very robust, particularly compared to the current proposal for advanced television. This is made possible by the allocation of most or all ofthe bits to the base layer when very stressful image material is encountered. Such stressful material is by its nature both noise-like and very rapidly changing. In these circumstances, the eye cannot see detail associated with the enhancement layer of resolution. Since the bits are applied to the base layer, the reproduced frames are substantially more accurate than the currently proposed advanced television system, which uses a single constant higher resolution.
Thus, this aspect ofthe inventive system optimizes both perceptual and coding efficiency, while providing maximum visual impact. This system provides a very clean image at a resolution and frame rate performance that had been considered by many to be impossible. It is believed that this aspect ofthe inventive system is likely to outperform the advanced television formats being proposed by at this time. In addition to this anticipated superior performance, the invention also provides the highly valuable features of temporal and resolution layering.
While the discussion above has used MPEG-2 in its examples, these and other aspects of the invention may be carried out using other compression systems. For example, the invention will work with any comparable standard that provides I, B, and P frames or equivalents, such as MPEG-1, MPEG-2, MPEG-4, H.263, and other compression systems (including wavelets and other non-DCT systems).
ENHANCEMENTS TO LAYERED COMPRESSION
Overview A number of enhancements to the embodiments described above may be made to handle a variety of video quality and compression problems. The following describes a number of such enhancements, most of which are preferably embodied as a set of tools which can be applied to the tasks of enhancing images and compressing such images. The tools can be combined by a content developer in various ways, as desired, to optimize the visual quality and compression efficiency of a compressed data stream, particularly a layered compressed data stream.
Enhancement Layer Motion Vectors and Gray Bias
The usual way that a resolution enhancement layer is coded using MPEG-type (e.g., MPEG-2, MPEG-4, or comparable systems) compression is to bias a difference picture with a gray bias. In the common 8-bit pixel value range of 0=black to
255=white, the half-way point of 128 is commonly used as the gray bias value. Values below 128 represent negative differences between images, and values above 128 represent positive differences between the images. (For a 10-bit system, gray would be 512, 0=black, and 1023=white, and so forth for other bit ranges). The difference picture is found by subtracting an expanded and decompressed base layer from an original high resolution image. Sequences of these difference pictures are then encoded as an MPEG-type difference picture stream of frames, which operates as a normal MPEG-type picture stream. The gray bias value is removed when each difference picture is added to another image (for example, the expanded decoded base layer) to create an improved resolution result.
Instead of using motion vectors found on this difference picture stream, which is often quite noisy, it is usually preferable to find motion vectors on the original high resolution image. These motion vectors are then used to displace the difference pictures within each frame. Such motion vectors will track details better than motion vectors which are found on the difference pictures.
Each difference picture represents a delta adjustment which would be needed to make a perfect encoding ofthe high resolution original. The pixel difference delta values can only extend for half of the range, which is nearly always sufficient, since differences are usually quite small. Thus, black (typically 0) regions can be extended at most to half gray (at 127, typically), and white (typically 255) can be extended down at most to half gray (at 128, typically).
However, if a region ofthe original image were to be exactly mid-gray (at 128, typically), then the difference picture could be used to create an entire black-to-white range (of 0 to 255, typically).
This simple relationship can be utilized to widen the aspect ratio of a final image in addition to enhancing the resolution ofthe base layer. By expanding the decompressed base layer, and then extending the image to a larger width and/or height using mid-gray, the system can provide the gray value (128, typically) for use with a difference picture in order to add full picture detail outside ofthe extent ofthe expanded, decompressed base layer.
FIG. 11 is diagram of a base layer expanded by using a gray area and enhancement to provide picture detail. In particular, a base layer 1100 having a narrower aspect ratio is upfiltered and then expanded in area as an expanded base region 1102. The expanded base region 1102 is then "padded" with a uniform mid- gray pixel value (e.g., 128) to widen its aspect ratio or otherwise increase its size (an "additional area region" 1104). An enhancement layer can then be added having a small range of possible pixel values (i.e., a difference picture) for the area that coincides with the the expanded base region 1102, but a full range of possible pixel values (e.g., ±127) over the additional area region 1104, thus providing additional actual picture information.
In this way, the base layer can represent a narrow or shorter (or both) image extent than the enhanced image at higher resolution. The enhancement layer then contains both gray-biased image difference pictures, corresponding to the extent ofthe expanded decompressed base layer (i.e., the expanded base region 1102), as well as containing actual picture. Since the compressed enhancement layer is encoded as a standard MPEG-type picture stream, the fact that the edge region is actual picture, and the inner region is a difference picture, is not distinguished, and both are coded and carried along together in the same picture stream of frames.
In the preferred embodiment, the edge region, outside the expanded decompressed base layer extent, is a normal high resolution MPEG- type encoded stream. It has an efficiency which corresponds to normal MPEG- type encoding of a high resolution picture. However, since it is an edge region, motion vectors within the difference picture region should be constrained not to point into the border region (which contains actual picture information). Also, motion vectors for the border actual-picture region should be constrained not to point into the inner difference picture region. In this way the border actual-picture region coding and difference picture region coding will be naturally separated.
This can be accomplished best by finding all ofthe motion vectors on the original image, but constrain them not to cross the boundary between the inner difference picture region and the outer border actual-picture region. This is best done if macroblock boundaries fall on the boundary between the inner difference picture region and the outer border actual-picture region. Otherwise, if the difference picture edge with the actual picture border is in the central region of a macroblock, then additional bits will need to be used when coding to accomplish the transition to and from difference and actual picture border regions. Accordingly, greatest efficiency is obtained if macroblock boundaries are on the same edge as the edge between the inner difference picture region and the outer border actual-picture region. Note that the quantizer and rate buffer control during encoding of these hybrid difference-plus-actual-picture image-expanding enhancement layer pictures may need special adjustment to differentiate the larger extent of signals in the border actual- picture region over the inner difference picture region. There is a tradeoff in the use of this technique concerning the amount ofthe extent ofthe border actual-picture region. For small border extensions, the number of bits in proportion to the overall stream is small, but the relative efficiency for the small area is reduced because ofthe number of motion vectors which cannot find matches since such matches would be off the edge ofthe border region. Another way of looking at this is that the border region has a high proportion of edge to area, unlike a usual image rectangle, which has a much lower proportion of edge to area. The inner rectangular picture region, typical of normal digital video as is usually coded with compression such as MPEG-2 or MPEG-4, has a high degree of matches when finding motion vectors since most ofthe area within the frame, except at the very frame edges, is usually present in the previous frame. On a pan, for example, the direction of picture coming on-screen will cause one edge to have to create picture from nothing, since the image is coming from off-screen for each frame. However, most of a normal picture rectangle is on-screen in the previous frame, allowing the motion vectors to most often find matches. However, using this inventive border extension technique, the border area has a much higher percentage of off-screen mismatches in previous frames for motion compensation, since the screen outer edge, as well as the difference picture inner edge, are both "out-of-bounds" for motion vectors. Thus, some loss of efficiency is inherent in this approach when considered as bits per image area (or per pixel or per macroblock, which are equivalent bits-per-area measures). Thus, when the border region is relatively small, this relative inefficiency is a sufficiently small proportion of the overall bit rate to be acceptable. If the border is relatively large, likewise, the efficiency becomes higher, and the proportion may again be acceptable. Moderately sized borders may suffer some inefficiency during pans, but this inefficiency may be acceptable. One way that efficiency may be regained using this technique is that simpler ratios of base to enhancement layer resolution, such as 3/2, 4/3, and especially exact factors of 2, can be used for narrower base layers. Using factors of two, especially, helps gain significant efficiency in overall encoding using a base and resolution enhancement layer.
The lower resolution image may also be most naturally used on narrower screens, while the higher resolution image may be more naturally viewed on larger and wider, and/or taller screens.
It is also possible to continuously move or reposition the inner difference picture region, corresponding to the practice of "pan and scan" for the base resolution image. The upper borders would then have a dynamic re-position and size and shape. The macroblock alignment would usually be lost in continuous panning, but could be maintained if some care is taken to align cuts within the larger area. The simplest and most efficient construction, however, is a fixed-position centered alignment ofthe inner difference picture with respect to the base layer on exact macroblock boundaries.
Image Filtering
Downsizing and Upsizing Filters
Experimentation has shown that the downsizing filter used in creating a base layer from a high resolution original picture is most optimal if it includes modest negative lobes and an extent which stops after the first very small positive lobes after the negative lobes. FIG. 12 is a diagram ofthe relative shape, amplitudes, and lobe polarity of a preferred downsizing filter. The down filter essentially is a center- weighted function which has been truncated to a center positive lobe 1200, a symmetric pair of adjacent (bracketing) small negative lobes 1204, and a symmetric pair of adjacent (bracketing) very small outer positive lobes 1206. The absolute amplitude ofthe lobes 1200, 1202, 1204 may be adjusted as desired, so long as the relative polarity and amplitude inequality relationships shown in FIG. 12 are maintained. However, a good first approximation for the relative amplitudes are defined by a truncated sine function (sinc(x) = sin(x)/x). Such filters can be used separably, which means that the horizontal data dimension is independently filtered and resized, and then the vertical data dimension, or vise versa; the result is the same.
When creating a base layer original (as input to the base layer compression) from a low-noise high resolution original input, the preferred downsizing filter has first negative lobes which are of a normal sine function amplitude. For clean and for high resolution input images, this normal truncated sine function works well. For lower resolutions (e.g., 1280x720, 1024x768, or 1536x768), and for noisier input pictures, a reduced first negative lobe amplitude in the filters is more optimal. A suitable amplitude in such cases is about half the truncated sine function negative lobe amplitude. The small first positive lobes outside ofthe first negative lobes are also reduced to lower amplitude, typically to 1/2 to 2/3 ofthe normal sine function amplitude. The affect of reducing the first negative lobes is the main issue, since the small outside positive lobes do not contribute to picture noise. Further samples outside the first positive lobes preferably are truncated to minimize ringing and other potential artifacts.
The choice of whether to use milder negative lobes or full sine function amplitude negative lobes in the downfilter is determined by ofthe resolution and noise level ofthe original image. It is also somewhat a function of image content, since some types of scenes are easier to code than others (mainly related to the amount of motion and change in a particular shot). By using a "milder" downfilter having reduced negative lobes, noise in the base layer is reduced, and a cleaner and quieter compression ofthe base layer is achieved, thus also resulting in fewer artifacts.
Experimentation has also shown that the optimal upsizing filter has a center positive lobe with small adjacent negative lobes, but no further positive lobes. FIGS. 13 A and 13B are diagrams ofthe relative shape, amplitudes, and lobe polarity of a pair of preferred upsizing filters for upsizing by a factor of 2. A central positive lobe 1300, 1300' is bracketed by a pair of small negative lobes 1302, 1302'. An asymmetrically placed positive lobe 1304, 1304' is also required. These paired upfilters could also be considered to be truncated sine filters centered on the newly created samples. For example, for a factor of two upfilter, two new samples will be created for each original sample. The small adjacent negative lobes 1302, 1302' have less negative amplitude than is used in the corresponding downsizing filter (FIG. 12), or than would be used in an optimal (sine-based) upsizing filter for normal images. This is because the images being upsized are decompressed, and the compression process changes the spectral distribution. Thus, more modest negative lobes, and no additional positive lobes beyond the middle ones 1300, 1300', work better for upsizing a decompressed base layer.
Experimentation has shown that slight negative lobes 1302, 1302' provide a better layered result than positive-only gaussian or spline upfilters (note that splines can have negative lobes, but are most often used in the positive-only form). Thus, this upsizing filter preferably is used for the base layer in both the encoder and the decoder.
Weighting of High Octave of Picture Detail
In the preferred embodiment, the signal path which expands the original uncompressed base layer input image uses a gaussian upfilter rather than the upfilter described above. In particular, a gaussian upfilter is used for the "high octave" of picture detail, which is determined by subtracting the expanded original base- resolution input image (without using compression) from the original picture. Thus, no negative lobes are used for this particular upfiltered expansion.
As noted above, for MPEG-2 this high octave difference signal path is typically weighted with 0.25 (or 25%) and added to the expanded decompressed base layer (using the other upfilter described above) as input to the enhancement layer compression process. However, experimentation has shown that weights of 10%, 15%, 20%, 30%, and 35% are useful for particular images when using MPEG-2. Other weights may also prove useful. For MPEG-4, it has been found that filter weights of 4-8% may be optimal when used in conjunction with other improvements described below. Accordingly, this weighting should be regarded as an adjustable parameter, depending upon the encoding system, the scenes being encoded/compressed, the particular camera (or film) being used, and the image resolution. De-Interlacing and Noise Reduction Enhancements
Overview
Experimentation has shown that many de-interlacing algorithms and devices depend upon the human eye to integrate fields to create an acceptable result. However, since compression algorithms are not a human eye, any integration of de-interlaced fields should take into account the characteristics of such algorithms. Without such careful de-interlaced integration, the compression process will create high levels of noise artifacts, both wasting bits (hindering compression) as well as making the image look noisy and busy with artifacts. This distinction between de-interlacing for viewing (such as with line-doublers and line-quadruplers) vs. de-interlacing as input to compression, has lead to the techniques described below. In particular, the de- interlacing techniques described below are useful as input to single-layer noninterlaced MPEG-like, as well as to the layered MPEG-like compression described above. Further, noise reduction must similarly match the needs of being an input to compression algorithms, rather than just reducing noise appearance. The goal is generally to reproduce, upon decompression, no more noise than the original camera or film-grain noise. Equal noise is generally considered acceptable, after compression/decompression. Reduced noise, with equivalent sharpness and clarity with the original, is a bonus. The noise reduction described below achieves these goals.
Further, for very noisy shots, such as from high speed film or with high camera sensitivity settings, usually in low light, noise reduction can be the difference between a good looking compressed/decompressed image vs. one which is unwatchably noisy. The compression process greatly amplifies noise which is above some threshold of acceptability to the compressor. Thus, the use of noise-reduction pre-processing to keep noise below this threshold may be required for acceptable good quality results. De-Graining and Noise-Reducing Filters
It has been found through experimentation that applying de-graining and/or noise-reducing filtering before layered or non-layered encoding improves the ability of the compression system to perform. While de-graining or noise-reduction is most effective on grainy or noisy images prior to compression, either process may be helpful when used in moderation even on relatively low noise or low grain pictures. Any of several known de-graining or noise-reduction algorithms may be applied. Examples are "coring", simple neighbor median filters, and softening filters.
Whether noise-reduction is needed is determined by how noisy the original images are. For interlaced original images, the interlace itself is a form of noise, which usually will require additional noise reduction filtering, in addition to the complex de- interlacing process described below. For progressive scan (non-interlaced) camera or film images, noise processing is useful in layered and non-layered compression when noise is present above a certain level. There are different types of noise. For example, video transfers from film include film grain noise. Film grain noise is caused by silver grains which couple to yellow, cyan, and magenta film dyes. Yellow affects both red and green, cyan affects both blue and green, and magenta affects both red and blue. Red is formed where yellow and magenta dye crystals overlap. Similarly green is the overlap of yellow and cyan, and blue is the overlap of magenta and cyan. Thus, noise between colors is partially correlated through the dyes and grains between pairs of colors. Further, when multiple grains overlap in all three colors, as they do in a print dark regions ofthe image or on a negative in light regions ofthe image (dark on the negative), additional color combinations occur. This correlation between the colors can be utilized in film- grain noise reduction, but is a complex process. Further, many different film types are used, and each type has different grain sizes, shapes, and statistical distributions.
For video images created by CCD-sensor and other (e.g., tube) sensor cameras, the red, green, and blue noise is uncorrelated. In this case, it is best to process the red, green, and blue records independently. Thus, red noise is reduced with self-red processing independently of green noise and blue noise; the same approach applies to green and blue noise. Thus, noise processing is best matched to the characteristics ofthe noise source itself. In the case of a composite image (from multiple sources), the noise may differ in characteristics over different portions ofthe image. In this situation, generic noise processing may be the only option, if noise processing is needed. It has also been found useful in some cases to perform a "re-graining" or "re- noising" process after decoding a compressed layered data stream, as a creative effect, since some de-grained or de-noised images may be "too clean" or "too sterile" in appearance. Re-graining and/or re-noising are relatively easy effects to add in the decoder using any of several known algorithms. For example, this can be accomplished by the addition of low pass filtered random noise of suitable amplitude.
De-Interlacing Before Compression
As mentioned above, the preferred compression method for interlaced source which is ultimately intended for non-interlaced display includes a step to de-interlace the interlaced source before the compression steps. De-interlacing a signal after decoding in the receiver, where the signal has been compressed in the interlaced mode, is both more costly and less efficient than de-interlacing prior to compression, and then sending a non-interlaced compressed signal. The non-interlaced compressed signal can be either layered or non-layered (i.e., a conventional single layer compression). Experimentation has shown that filtering a single field of an interlaced source, and using that field as if it were a non-interlaced full frame, gives poor and noisy compression results. Thus, using a single-field de-interlacer prior to compression is not a good approach. Instead, experimentation has shown that a three-field-frame de- interlacer process using field synthesized frames ("field-frames"), with weights of [0.25, 0.5, 0.25] for the previous, current, and next field-frames, respectively, provides a good input for compression. Combining three field-frames may be performed using other weights (although these weights are optimal) to create a de-interlaced input to a compression process.
In the preferred de-interlacing system, a field-de-interlacer is used as the first step in the overall process to create field-frames. In particular, each field is de- interlaced, creating a synthesized frame where the total number of lines in the frame is derived from the half number of lines in a field. Thus, for example, an interlaced 1080 line image will have 540 lines per even and odd field, each field representing l/60th of a second. Normally, the even and odd fields of 540 lines will be interlaced to create 1080 lines for each frame, which represents l/30th of a second. However, in the preferred embodiment, the de-interlacer copies each scanline without modification from a specified field (e.g., the odd fields) to a buffer that will hold some ofthe de- interlaced result. The remaining intermediate scanlines (in this example, the even scanlines) for the frame are synthesized by adding half of the field line above and half ofthe field line below each newly stored line. For example, the pixel values of line 2 for a frame would each comprise 1/2 ofthe summed corresponding pixel values from each of line 1 and line 3. The generation of intermediate synthesized scanlines may be done on the fly, or may be computed after all ofthe scanlines from a field are stored in a buffer. The same process is repeated for the next field, although the field types (i.e., even, odd) will be reversed.
FIG. 14A is a block diagram of an odd-field de-interlacer, showing that the odd lines from an odd field 1400 are simply copied to a de-interlaced odd field 1402, while the even lines are created by averaging adjacent odd lines from the original odd field together to form the even lines ofthe de-interlaced odd field 1402. Similarly, FIG. 14B is a block diagram of an even-field de-interlacer, showing that the even lines from an even field 1404 are simply copied to a de-interlaced even field 1406, while the odd lines are created by averaging adjacent even lines from the original even field together to form the odd lines ofthe de-interlaced even field 1406. Note that this case corresponde to "top field first"; "bottom field first" could also be considered the "even" field.
As a next step, a sequence of these de-interlaced fields is then used as input to a three-field-frame de-interlacer to create a final de-interlaced frame. FIG. 15 is a block diagram showing how the pixels of each output frame are composed of 25% of the corresponding pixels from a previous de-interlaced field (field-frame) 1502, 50% ofthe corresponding pixels from a current field-frame 1504, and 25% ofthe corresponding pixels from the next field-frame 1506. The new de-interlaced frame then contains much fewer interlace difference artifacts between frames than do the three field-frames of which it is composed. However, there is a temporal smearing by adding the previous field-frame and next field-frame into a current field-frame. This temporal smearing is usually not objectionable, especially in light ofthe de-interlacing improvements which result.
This de-interlacing process is very beneficial as input to compression, either single layer (unlayered) or layered. It is also beneficial just as a treatment for interlaced video for presentation, viewing, or making still frames, independent of use with compression. The picture from the de-interlacing process appears "clearer" than the presentation ofthe interlace directly, or ofthe de-interlaced fields.
De-Interlace Thresholding
Although the de-interlace three-field sum weightings of [0.25, 0.5, 0.25] discussed above provide a stable image, moving parts of a scene can sometimes become soft or can exhibit aliasing artifacts. To counteract this, a threshold test may be applied which compares the result ofthe [0.25, 0.5, 0.25] temporal filter against the corresponding pixel values of only the middle field-frame. If a middle field-frame pixel value differs more than a specified threshold amount from the value ofthe corresponding pixel from the three-field-frame temporal filter, then only the middle field-frame pixel value is used. In this way, a pixel from the three-field-frame temporal filter is selected where it differs less than the threshold amount from the corresponding pixel ofthe single de-interlaced middle field-frame, and the middle field-frame pixel value is used when there is more difference than the threshold. This allows fast motion to be tracked at the field rate, and smoother parts ofthe image to be filtered and smoothed by the three-field-frame temporal filter. This combination has proven an effective, if not optimal, input to compression. It is also very effective for processing for direct viewing to de-interlace image material (also called line doubling in conjunction with display).
The preferred embodiment for such threshold determinations uses the following equations for corresponding RGB color values from the middle (single) deinterlaced field-frame image and the three-field-frame deinterlaced image: Rdiff = R_single_field_deinterlaced minus R_three_field_deinterlaced Gdiff = G_single_field_deinterlaced minus G_three_field_deinterlaced Bdiff = B_single_field_deinterlaced minus B_three_field_deinterlaced ThresholdingValue = abs(Rdiff+Gdiff+Bdiff) + abs(Rdiff) + abs(Gdiff)+ abs(Bdiff)
The ThresholdingValue is then compared to a threshold setting. Typical threshold settings are in the range of 0.1 to 0.3, with 0.2 being most common.
In order to remove noise from this threshold, smooth-filtering the three-field- frame and single-field-frame de-interlaced pictures can be used before comparing and thresholding them. This smooth filtering can be accomplished simply by down filtering (e.g., down filtering by two using the preferably down filter described above), and then up filtering (e.g., using a gaussian up-filter by two. This "down-up" smoothed filter can be applied to both the single-field-frame de-interlaced picture and the three-field-frame de-interlaced picture. The smoothed single-field-frame and three-field-frame pictures can then compared to compute a ThresholdingValue and then thresholded to determine which picture will source each final output pixel.
In particular, the threshold test is used as a switch to select between the single- field-frame de-interlaced picture and the three-field-frame temporal filter combination of single-field-frame de-interlaced pictures. This selection then results in an image where the pixels are from the three-field-frame de-interlacer in those areas where that image differs in small amounts (i.e., below the threshold) from the single field-frame image, and where the pixels are from the single field-frame image in those areas where the three-field-frame differed more than then the threshold amount from the single-field-frame de-interlaced pixels (after smoothing). This technique has proven effective in preserving single-field fast motion details (by switching to the single-field-frame de-interlaced pixels), while smoothing large portions ofthe image (by switching to the three-field-frame de-interlaced temporal filter combination).
In addition to selecting between the single-field-frame and three-field-frame de-interlaced image, it is also often beneficial to add a bit ofthe single-field-frame image to the three-field-frame de-interlaced picture, to preserve some ofthe immediacy ofthe single field pictures over the entire image. This immediacy is balanced against the temporal smoothness ofthe three-field-frame filter. A typical blending is to create new frame by adding 33.33% (1/3) of a single middle field-frame to 66.67% (2/3) ofthe corresponding three-field-frame smoothed image. This can be done before or after threshold switching, since the result is the same either way, only affecting the smoothed three-field-frame picture. Note that this is effectively equivalent to using a different proportion ofthe three field-frames, rather than the original three-field-frame weights of [0.25, 0.5, 0.25]. Computing 2/3 of [0.25, 0.5, 0.25] plus 1/3 of (0,1,0), yields [0.1667, 0.6666, 0.1667] as the temporal filter for the three field-frames. The more heavily weighted center (current) field-frame brings additional immediacy to the result, even in the smoothed areas which fell below the threshold value. This combination has proven effective in balancing temporal smoothness with immediacy in the de-interlacing process for moving parts of a scene.
Use of Linear Filters
Sums, filters, or matrices involving video pictures should take into account the fact that pixel values in video are non-linear signals. For example, the video curve for HDTV can be several variations of coefficients and factors, but a typical formula is the international CCIR XA-11 (now called Rec. 709): V = 1.0993 * L° 45 - 0.0993 for L > 0.018051
V = 4.5 * L for L <= 0.018051 where V is the video value and L is linear light luminance.
The variations adjust the threshold (0.018051) a little, the factor (4.5) a little (e.g. 4.0), and the exponent (0.45) a little (e.g., 0.4). The fundamental formula, however, remains the same.
A matrix operation, such as a RGB to/from YUV conversion, implies linear values. The fact that MPEG in general uses the video non-linear values as if they were linear results in leakage between the luminance (Y) and the color values (U, and V). This leakage interferes with compression efficiency. The use of a logarithmic representation, such as is used with film density units, corrects much of this problem. The various types of MPEG encoding are neutral to the non-linear aspects ofthe signal, although its efficiency is effected due to the use ofthe matrix conversion RGB to/from YUV. YUN (U = R-Y, V = B-Y) should have Y computed as a linearized sum of 0.59 G, plus 0.29 R, plus 0.12 B (or slight variations on these coefficients). However, U (= R-Y) becomes equivalent to R/Y in logarithmic space, which is orthogonal to luminance. Thus, a shaded orange ball will not vary the U (= R-Y) parameter in a logarithmic representation. The brightness variation will be represented completely in the Luminance parameter, where full detail is provided.
The linear vs. logarithmic vs. video issue impacts filtering. A key point to note is that small signal excursions (e.g. 10% or less) are approximately correct when a non-linear video signal is processed as if it were a linear signal. This is because a piece-wise linear approximation to the smooth video-to-from-linear conversion curve is reasonable. However, for large excursions, a linear filter is much more effective, and produces much better image quality. Accordingly, if large excursions are to be optimally coded, transformed, or otherwise processed, it would be desirable to first convert the non-linear signal to a linear one in order to be able to apply a linear filter.
De-interlacing is therefore much better when each filter and summation step utilizes conversions to linear values prior to filtering or summing. This is due to the large signal excursions inherent in interlaced signals at small details ofthe image. After filtering, the image signals are converted back to the non-linear video digital representation. Thus, the three-field-frame weighting (e.g., [0.25, 0.5, 0.25] or [0.1667, 0.6666, 0.1667]) should be performed on a linearized video signal. Other filtering and weighted sums of partial terms in noise and de-interlace filtering should also be converted to linear form for computation. Which operations warrant linear processing is determined by signal excursion, and the type of filtering. Image sharpening can be appropriately computed in video or logarithmic non-linear representations, since it is self-proportional. However, matrix processing, spatial filtering, weighted sums, and de-interlace processing should be computed using linearized digital values. As a simple example, the single field-frame de-interlacer described above computes missing alternate lines by averaging the line above and below each actual line. This average is much more correct numerically and visually if this average is done linearly. Thus, instead of summing 0.5 times the line above plus 0.5 times the line below, the digital values are linearized first, then averaged, and then reconverted back into the non-linear video representation.
Layered Mode Based Upon 2/3 Base Layer
A 1280x720 enhancement layer can utilize an 864x480 base layer (i.e., a 2/3 relationship between the enhancement and base layer). FIG. 16 is a block diagram of such a mode. An original image 1600 at 1280x720 is padded to 1296x720 (to be a multiple of 16) and then downsized by 2/3 to an 864x480 image 1602 (also a multiple of 16). The downsizing preferably uses a normal filter or a filter having mild negative lobes. As described above, this downsized image 1602 may be input to an first encoder 1604 (e.g., an MPEG-2 or MPEG-4 encoder) for direct encoding as a base layer.
To encode the enhancement layer, the base layer is decompressed and upsized (expanded and up-filtered) by 3/2 to a 1296x720 intermediate frame 1606. The upfilter preferably has mild negative lobes. This intermediate frame 1606 is subtracted from the original image 1600. Meanwhile, the 864x480 image 1602 is up-filtered by 3/2 (preferably using a gaussian filter) to 1280x720 and subtracted from the original image 1600. The result is weighted (e.g., typically by 25% for MPEG-2) and added to the result ofthe subtraction ofthe intermediate frame 1606 from the original image 1600. This resulting sum is cropped to a reduced size (e.g., 1152x688) and the edges feathered, resulting in a pre-compression enhancement layer frame 1608. This pre- compression enhancement layer frame 1608 is applied as an input to a second encoder 1610 (e.g., an MPEG-2 or MPEG-4 encoder) for encoding as an enhancement layer. The efficiency and quality at 18.5 mbits/sec is approximately equivalent between "single" layered (i.e., non-layered) and a layered system using this configuration. The efficiency of a 2/3 relationship between the enhancement and base layer is not as good as when using a factor of two, since the DCT coefficients are less orthogonal between the base and enhancement layers. However, this construction is workable, and has the advantage of providing a high quality base layer (which is cheaper to decode). This is an improvement over the single layered configuration where the entire high resolution picture must be decoded (at a higher cost), when lower resolution is all that can be provided by a particular display.
The layered configuration also has the advantage that the enhancement sub- region is adjustable. Thus, efficiency can be controlled by adjusting the size ofthe enhancement layer and the proportion ofthe total bit rate that is allocated to the base layer vs. the enhancement layer. Adjustment ofthe enhancement layer size and bit proportion can be used to optimize compression performance, especially under high stress (rapid motion or many scene changes). For example, as noted above, all ofthe bits may be allocated to the base layer under extreme stress.
Favorable resolution relationships between the enhancement and base layers are factors of 1/2, 2/3, and other simple fractions (e.g., 1/3, 3/4). It is also useful to apply squeezes with respect to the relationship between an enhancement layer and the base layer. For example, a source picture of 2048x1024 could have a base layer of 1536x512, which has a horizontal relationship of 3/4 and a vertical relationship of 1/2 with respect to the source image. Although this is not optimal (a factor of two both horizontally and vertically is optimal), it is illustrative ofthe principle. The use of 2/3 both horizontally and vertically might be improved upon for some resolutions by using a factor of 2 vertically and a factor of 2/3 horizontally. Alternatively, it may be more optimal for some resolutions to use a factor of 2/3 vertically and 1/2 horizontally. Thus, simple fractions such as 1/2, 2/3, 3/4, 1/3, etc. can be independently applied to the horizontal and vertical resolution relationships, allowing a large number of possible combinations of relationships. Thus, the relationship ofthe full input resolution to the resolution ofthe base layer, as well as the relationship of the enhancement layer to the base layer and the input resolution, allows full flexibility in the use of such fractional relationships. Particularly useful combinations of such resolution relationships may be assigned a compression "enhancement mode" number if adopted as part of any standard. Median Filters
In noise processing, the most useful filter is the median filter. A three element median filter just ranks the three entries, via a simple sort, and picks the middle one. For example, an X (horizontal) median filter looks at the red value (or green or blue) of three adjacent horizontal pixels, and picks the one with the middle-most value. If two are the same, that value is selected. Similarly, a Y filter looks in the scanlines above and below the current pixel, and again picks the middle value.
It has been experimentally determined that it is useful to average the results from applying both an X and a Y median filter to create a new noise-reducing component picture (i.e., each new pixel is the 50% equal average ofthe X and Y medians for the corresponding pixel from a source image).
In addition to X and Y (horizontal and vertical) medians, it is also possible to take diagonal and other medians. However, the vertical and horizontal pixel values are most close physically to any particular pixel, and therefore produce less potential error or distortion than the diagonals. However, such other medians remain available in cases where noise reduction is difficult using only the vertical and horizontal medians.
Another beneficial source of noise reduction is information from the previous and subsequent frame (i.e., a temporal median). As mentioned below, motion analysis provides the best match for moving regions. However, it is compute intensive. If a region ofthe image is not moving, or is moving slowly, the red values (and green and blue) from a current pixel can be median filtered with the red value at that same pixel location in the previous and subsequent frames. However, odd artifacts may occur if significant motion is present and such a temporal filter is used. Thus, it is preferred that a threshold be taken first, to determine whether such a median would differ more than a selected amount from the value of a current pixel. The threshold can be computed essentially the same as for the de-interlacing threshold above:
Rdiff= R_current_pixel minus R_temporal_median
Gdiff = G_current_pixel minus G_temporal_median
Bdiff = B_current_pixel minus B_temporal_median ThresholdingValue = abs(Rdiff+Gdiff+Bdiff) + abs(Rdiff) + abs(Gdiff)+ abs(Bdiff) The ThresholdingValue is then compared to a threshold setting. Typical threshold settings are in the range 0.1 to 0.3, with 0.2 being typical. Above the threshold, the current value is kept. Below the threshold, the temporal median is used.
An additional median type is a median taken between the X, Y, and temporal medians. Another median type can take the temporal median, and then take the equal average ofthe X and Y medians from it.
Each type of median can cause problems. X and Y medians smear and blur an image, so that it looks "greasy". Temporal medians cause smearing of motion over time. Since each median can result in problems, yet each median's properties are different (and, in some sense, "orthogonal"), it has been determined experimentally that the best results come by combining a variety of medians.
In particular, a preferred combination of medians is a linear weighted sum (see the discussion above on linear video processing) of five terms to determine the value for each pixel of a current image: 50% ofthe original image (thus, the most noise reduction is 3db, or half);
15% ofthe average of X and Y medians;
10% ofthe thresholded temporal median;
10%) ofthe average of X and Y medians ofthe thresholded temporal median; and 15% of a three-way X, Y, and temporal median.
This set of time medians does a reasonable job of reducing the noise in the image without making it appear "greasy" or blurred, causing temporal smearing of moving objects, or losing detail. Another useful weighting of these five terms is 35%, 20%, 22.5%, 10%, and 12.5%, respectively. In addition, it is useful to apply motion-compensation by applying center weighted temporal filters to a motion-compensated mm region, as described below. This can be added to the median filtered image result (of five terms, just described) to further smooth the image, providing better smoothing and detail on moving image regions. Motion Analysis
In addition to "in-place" temporal filtering, which does a good job at smoothing slow-moving details, de-interlacing and noise reduction can also be improved by use of motion analysis. Adding the pixels at the same location in three fields or three frames is valid for stationary objects. However, for moving objects, if temporal averaging/smoothing is desired, it is often more optimal to attempt to analyze prevailing motion over a small group of pixels. For example, an nxn block of pixels (e.g., 2x2, 3x3, 4x4, 6x6, or 8x8) can be used to search in previous and subsequent fields or frames to attempt to find a match (in the same way MPEG-2 motion vectors are found by matching 16x16 macroblocks). Once a best match is found in one or more previous and subsequent frames, a "trajectory" and "moving mini-picture" can be determined. For interlaced fields, it is best to analyze comparisons as well as compute inferred moving mini-pictures utilizing the results of the thresholded de-interlaced process above. Since this process has already separated the fast-moving from the slow-moving details, and has already smoothed the slow moving details, the picture comparisons and reconstructions are more applicable than individual de-interlaced fields.
The motion analysis preferably is performed by comparison of an nxn block in the current thresholded de-interlaced image with all nearby blocks in the previous and subsequent one or more frames. The comparison may be the absolute value of differences in luminance or RGB over the nxn block. One frame is sufficient forward and backward if the motion vectors are nearly equal and opposite. However, if the motion vectors are not nearly equal and opposite, then an additional one or two frames forward and backward can help determine the actual trajectory. Further, different de- interlacing treatments may be useful in helping determine the "best guess" motion vectors going forward and back. One de-interlacing treatment can be to use only individual de-interlaced fields, although this is heavily prone to aliasing and artifacts on small moving details. Another de-interlacing technique is to use only the three- field-frame smooth de-interlacing, without thresholding, having weightings [0.25, 0.5, 0.25], as described above. Although details are smoothed and sometimes lost, the trajectory may often be more correct. Once a trajectory is found, a "smoothed nxn block" can be created by temporally filtering using the motion-vector-offset pixels from the one (or more) previous and subsequent frames. A typical filter might again be [0.25, 0.5, 0.25] or [0.1667, 0.6666, 0.1667] for three frames, and possibly [0.1, 0.2, 0.4, 0.2, 0.1] for two frames back and forward. Other filters, with less central weight, are also useful, especially with smaller block sizes (such as 2x2, 3x3, and 4x4). Reliability ofthe match between frames is indicated by the absolute difference value. Large minimum absolute differences can be used to select more center weight in the filter. Lower values of absolute differences can suggest a good match, and can be used to select less center weight to more evenly distribute the average over a span of several frames of motion-compensated blocks.
These filter weights can be applied to: individual de-interlaced motion- compensated field-frames; thresholded three-field-frame de-interlaced pictures, described above; and non-thresholded three-field-frame de-interlaced images, with a [0.25, 0.5, 0.25] weighting, also as described above. However, the best filter weights usually come from applying the motion-compensated block linear filtering to the thresholded three-field-frame result described above. This is because the thresholded three-field-frame image is both the smoothest (in terms of removing aliasing in smooth areas), as well as the most motion-responsive (in terms of defaulting to a single de-interlaced field-frame above the threshold). Thus, the motion vectors from motion analysis can be used as the inputs to multi-frame or multi-de-interlaced-field- frame or single-de-interlaced field-frame filters, or combinations thereof. The thresholded multi-field-frame de-interlaced images, however, form the best filter input in most cases. The use of motion analysis is computationally expensive for a large search region, when fast motion might be found (such as ±32 pixels). Accordingly, it may be best to augment the speed by using special-purpose hardware or a digital signal processor assisted computer.
Once motion vectors are found, together with their absolute difference measure of accuracy, they can be utilized for the complex process of attempting frame rate conversion. However, occlusion issues (objects obscuring or revealing others) will confound matches, and cannot be accurately inferred automatically. Occlusion can also involve temporal aliasing, as can normal image temporal undersampling and its beat with natural image frequencies (such as the "backward wagon wheel" effect in movies). These problems often cannot be unraveled by any known computation technique, and to date require human assistance. Thus, human scrutiny and adjustment, when real-time automatic processing is not required, can be used for offline and non-real-time frame-rate conversion and other similar temporal processes.
De-interlacing is a simple form ofthe same problem. Just as with frame-rate- conversion, the task of de-interlacing is theoretically impossible to perform perfectly. This is especially due to the temporal undersampling (closed shutter), and an inappropriate temporal sample filter (i.e., a box filter). However, even with correct samples, issues such as occlusion and interlace aliasing further ensure the theoretical impossibility of correct results. The cases where this is visible are mitigated by the depth ofthe tools, as described here, which are applied to the problem. Pathological cases will always exist in real image sequences. The goal can only be to reduce the frequency and level of impairment when these sequences are encountered. However, in many cases, the de-interlacing process can be acceptably fully automated, and can run unassisted in real-time. Even so, there are many parameters which can often benefit from manual adjustment.
Filter Smoothing of High Frequencies
In addition to median filtering, reducing high frequency detail will also reduce high frequency noise. However, this smoothing comes at the price of loss of sharpness and detail. Thus, only a small amount of such smoothing is generally useful. A filter which creates smoothing can be easily made, as with the threshold for de-interlacing, by down-filtering with a normal filter (e.g., truncated sine filter) and then up-filtering with a gaussian filter. The result will be smoothed because it is devoid of high frequency picture detail. When such a term is added, it typically must be in very small amounts, such as 5% to 10%, in order to provide a small amount of noise reduction. In larger amounts, the blurring effect generally becomes quite visible. Base Layer Noise Filtering
The filter parameters for the median filtering described above for an original image should be matched to the noise characteristics ofthe film grain or image sensor that captured the image. After this median filtered image is down-filtered to generate an input to the base layer compression process, it still contains a small amount of noise. This noise may be further reduced by a combination of another X-Y median filters (equally averaging the X and Y medians), plus a very small amount ofthe high frequency smoothing filter. A preferred filter weighting of these three terms, applied to each pixel ofthe base layer, is: 70% ofthe original base layer (down filtered from median-filtered original above);
22.5% ofthe average of X and Y medians; and
7.5% ofthe down-up smoothing filter.
This small amount of additional filtering in the base layer provides a small additional amount of noise reduction and improved stability, resulting in better MPEG encoding and limiting the amount of noise added by such encoding.
Filters with Negative Lobes For Motion Compensation in MPEG-2 and MPEG-4
In MPEG-4, reference filters have been implemented for shifting macroblocks when finding the best motion vector match, and then using the matched region for motion compensation. MPEG-4 video coding, like MPEG-2, supports 1/2 pixel resolution of motion vectors for macroblocks. Unlike MPEG-2, MPEG-4 also supports 1/4 pixel accuracy. However, in the reference implementation of MPEG-4, the filters used are sub-optimal. In MPEG-2, the half-way point between pixels is just the average ofthe two neighbors, which is a sub-optimal box filter. In MPEG-4, this filter is used for 1/2 pixel resolution. If 1/4 pixel resolution is invoked in MPEG-4 Version2, a filter with negative lobes is used for the half-way point, but a sub-optimal box filter with this result and the neighboring pixels is used for the 1/4 and 3/4 points. Further, the chrominance channels (U = R-Y and V=B-Y) do not use any sub- pixel resolution in the motion compensation step under MPEG-4. Since the luminance channel (Y) has resolution to the 1/2 or 1/4 pixel, the half-resolution chrominance U and V channels should be sampled using filters to 1/4 pixel resolution, corresponding to 1/2 pixel in luminance. When 1/4 pixel resolution is selected for luminance, then 1/8 pixel resolution should be used for U and V chrominance.
Experiments have shown that the effects of filtering are significantly improved by using a negative lobe truncated sine function (as described above) for filtering the 1/4, 1/2, and 3/4 pixel points when doing 1/4 pixel resolution in luminance, and by using similar negative lobes when doing 1/2 pixel resolution for the filter which creates the 1/2 pixel position.
Similarly, effects of filtering are significantly improved by using a negative lobe truncated sine function for filtering the 1/8-pixel points for U and V chrominance when using 1/4 pixel luminance resolution, and by using 1/4 pixel resolution filters with similar negative lobe filters when using 1/2 pixel luminance resolution.
It has been discovered that the combination of quarter-pixel motion vectors with truncated sine motion compensated displacement filtering results in a major improvement in picture quality. In particular, clarity is improved, noise and artifacts are reduced, and chroma detail is increased.
These filters may be applied to video images under MPEG-1, MPEG-2, MPEG-4 or any other appropriate motion-compensated block-based image coding system.
Imaging Device Characterization and Correction
In working with particular progressive-scan (non-interlaced) cameras, it has been experimentally determined that it is highly desirable to apply pre-processing specific to a particular camera prior to compression (either layered or non-layered). For example, in one camera type, there is a mechanical horizontal misalignment of 1/3 of a pixel between the sensors for red and green, and another 1/3 pixel between the sensors for green and blue (2/3 pixel between red and blue). This results in color fringes around tiny vertical details. These color fringes, although not visible to the eye in the original image, result in color noise in the compression/decompression process which is very visible and objectionable. A pre-process specific to this one camera type corrects this color displacement, resulting in an input to the compression which then does not have color artifacts. Thus, although invisible, such tiny nuances in the properties of cameras and their sensors become critical to the acceptability and quality ofthe final compressed/decompressed results. Thus, it is useful to distinguish between "what the eye sees" vs. "what the compressor sees". This distinction has been used to advantage to discover preprocessing steps which greatly improve the quality of a compressed/decompressed image.
Accordingly, each individual electronic camera, each camera type, each film- type, and each individual film scanner and scanner type used in creating input to a compression/decompression system should be individually characterized in terms of color alignment and noise (electronic noise for video cameras and scanners, and grain for film). The information about where the image was created, a table ofthe specific properties, and specific settings of each piece of equipment, should be carried with the original image, and subsequently used in pre-processing prior to compression.
For example, a specific camera may require a color realignment. It may also be set on a medium noise setting (substantially affecting the amount of noise processing needed). These camera settings and inherent camera properties should be carried as side information along with each shot from that camera. This information then can be used to control the type of pre-processing, and the settings of parameters for the pre- processes. For images which are edited from multiple cameras, or even composited from multiple cameras and/or film sources, the pre-processing should probably be performed prior to such editing and combining. Such pre-processing should not degrade image quality, and may even be invisible to the eye, but does have a major impact on the quality ofthe compression.
Following is a general methodology for performing and using such characterizations for non-film imaging systems (e.g., electronic cameras and film scanners) used to create images to be input into a particular compression system:
(1) Image a resolution test chart and measure the horizontal and vertical color alignment ofthe pixel sensors (grains, for film), by color pair (e.g., RG, RB,
GB), preferably expressed in pixel units.
(2) Image one or more monochrome test charts and measure the noise generated by the sensors individually and as a set (e.g., by imaging a white card, black card, 50% and 18% gray cards, and each of red, green, and blue reference cards), preferably expressed as red, green, and blue pixel values. Determine if the noise is correlated by comparing ouφut variations from other color channels and adjacent neighbor pixels.
(3) Convey the measured information created by the measured device along with the image (e.g., by electronic transmission, storage on a machine readable medium, or by human-readable data accompanying the image).
(4) Before using images from the imaging system in a compression process, translate the pixels, by color, by an equal offset amount to correct for any measured misalignment. For example, if the red sensor is misaligned 0.25 pixels below the blue sensor, then all red pixels in an image should be shifted upwards by 0.25 pixels. Similarly, based on the measured amounts of noise, adjust the noise reduction filter weights by an amount that compensates for the amount of measured noise (this may need to be empirically determined and defined in a manual or computerized look-up table).
Following is a general methodology for performing and using such characterizations for film imaging systems used to create images to be input into a particular compression system: (1) Determine the film type (grain varies by film type).
(2) Expose the film to one or more monochrome test charts under a variety of lighting conditions (noise is in part a function of exposure).
(3) Scan the film at normal speed through a film scanner (whose characteristics are measured as above) and measure the noise generated by the sensors individually and as a set. Determine if the noise is correlated.
(4) Whenever film ofthe same type is exposed and then scanned by the measured scanner, convey the determined and measured information (i.e., film type, exposure conditions, scanning characteristics) along with the scanned film image. (5) Before using such images in a compression process, adjust the noise reduction filter weights by an amount that compensates for the amount of measured noise (this may need to be empirically determined and defined in a manual or computerized look-up table; a computer is preferable because the adjustment may be a function of at least three factors: film type, exposure conditions, and scanning characteristics) .
Enhanced 3-2 Pulldown System
It is a common and highly disliked practice for film to be transferred to 60Hz video using the 3-2 pulldown method described above. The 3-2 pulldown method is used because 24 frames per second do not divide evenly into 59.94 or 60 fields per second for existing NTSC (and some proposed HDTV) systems. The odd frames (or even) are placed on two ofthe interlaced fields, and the even frames (or odd) are placed on three ofthe interlaced fields. Thus, one field is a duplicate in every five fields. One frame of film maps to five fields of video. As noted above, this process leads to numerous unpleasant problems. Most video processing equipment only applies its process to an immediate signal. With this being the case, a time-changing effect will operate differently on one field than the next, even though some ofthe input fields were duplicates. After such processes, the fields are no longer duplicates, nor can field pairs be recombined to recover the original film frames. Examples of such processes, occurring at the field rate, include pan-and-scan (to move narrow 4:3 video screens horizontally across widescreen images to show important action), fade up or down, gradual color adjust, video title overlay scroll, etc. Further, if such a signal is captured on film, and then edited and processed on video, the frame processing ofthe film, and the field processing ofthe video, are horribly intermixed in an inextricable manner. When such a video signal (which occur widely) is then fed to an image compression system, the system generally performs sub-optimally.
Experiments have shown that, to date, the best image compression from a film source occurs only when the 24 φs images ofthe film can be perfectly re-extracted from the video signal (or better yet, never leave the 24 φs realm). Then the compression system can code the movie (or film-based TV show or commercial) at the original 24 φs rate ofthe original film. This is the most efficient manner of compression. Some movie-on-demand systems and DVD mastering systems are careful to apply 3-2 pulldown and editing in very limited ways, to ensure that the 24 φs original frames can be finally extracted and compressed at 24 φs. However, such care is "open loop", and is often violated by normal human error. The complexity of editing and applying post-production effects to a production often leads to "mistakes" where field-rate processing occurs. Accordingly, a preferred methodology that avoids such a possibility and eliminates the complexity of attempting to keep track of everything so as to avoid such errors, is as follows: (1) Whenever possible, utilize equipment for film processing which supports direct 24 φs storage, processing, or communication.
(2) Use electronic or fast optical media (e.g., hard drive and/or RAM) for local storage, and store all film images at their native 24 φs rate.
(3) Whenever a device takes 3-2-pulldown video as an input, make the 3-2 pulldown on the fly (in real time) converted from local storage (which is kept at
24 φs).
(4) When storing the ouφut of any device which produces and communicates 3-2 pulldown images, undo the 3-2 pulldown on-the fly, and store again at 24 φs. (5) Eliminate all devices from the system which must operate only on fields such that frames cannot be preserved with common processing (for 2 and for 3 fields, as one frame).
(6) Set all software which performs effects or editing on the stored image sequences to match the 24 φs mode which is used on the storage media; use no software which cannot operate in 24 φs native mode.
(7) Telecine (i.e., convert from film to video) all original images with a deterministic cadence (i.e., either always 3 then 2, or 2 then 3) if the telecine does not provide direct 24 φs output. Undo the cadence immediately after the interlaced 3-2 pulldown interface from the telecine.
(8) If a tape is received with an unknown 3-2 pulldown cadence, the cadence must be discovered by some method, and removed prior to storage. This can be done with hardware detecting systems, software detecting systems, or manually/visually. Unfortunately, no hardware detection systems are perfect, so manual visual inspection may always be required. (Current systems attempt to detect field misalignment. Such misalignment cannot presently be detected on black or white frames, or any constant value field of image brightness. Even with detectable misalignment, some detectors fail due to noise or algorithmic weaknesses.)
(9) Any tape storage ouφut from the facility requiring 3-2 pulldown will be stored in a known cadence which is maintained purely, and not disturbed for the entire running time ofthe program.
By this methodology, any particular processing device requiring 3-2 pulldown as input and ouφut will get its input(s) made on the fly in real time from a 24 φs source. The cadence will always begin in a standard way for each input. The cadence ofthe device's output is then known, and must be identical to the cadence created on the fly as the devices' input. The cadence is then un-done by this a priori knowledge, and the frames are saved in 24 φs format on the storage medium.
This methodology requires real-time 3-2 pulldown undo and 3-2 pulldown synthesis. Unless the cadence comes from tape in an unknown format, the 24 φs nature ofthe frames will automatically be preserved by such a film-based telecine post-production system. The system will then automatically form an optimal input to compression systems (including layered compression process described above).
This process should be broadly useful in video and HDTV telecine facilities. Someday, when all devices accept a 24 φs (and other rate progressive scan) native signal input, ouφut, processing, and storage modes, such a methodology will no longer be needed. However, in the interim, many devices require 3-2 pulldown for the interface in and out, even though the devices have a targeted function to operate on film input. During this interim, the above methodology eliminates 3-2 pulldown problems and can be an essential element ofthe efficiency of post-production and telecine of film.
Frame Rate Methods for Production
Although 24 φs has formed a world-wide standard for motion picture film, the use of 24 φs results in jumpy motion in many cases (also called "stutter" due to the multiple repeat flashes of a frame before moving to the next). Higher frame rates are desirable to provide smoother motion, a clearer picture for moving objects, as well as allowing slow motion (by capturing the images at a high frame rate, but playing them at a slower rate). As noted above, the video rate in the U.S. of 60 φs (and 59.94 φs in broadcast video) is relatively incompatible with 24 φs. This creates problems when attempting to release a movie worldwide, since 50 Hz PAL and SECAM video systems are relatively incompatible with 60 φs NTSC video and 60 Hz-centric US HDTV.
U.S. patent application Serial No. 09/435,277 (entitled "System And Method For Motion Compensation and Frame Rate Conversion", filed 11/5/1999, and assigned to the assignee ofthe present invention) teaches techniques which can perform difficult frame rate conversions such as 60 Uz to/from 50 Hz as well as 60 Hz to/and/from 72 Hz. These techniques also provide de-interlacing, in addition to frame rate conversion.
The results in using the frame rate conversion techniques taught in such application to convert between nearby high frame rates, such as 60 Hz to/from 50 Hz, or 60 Hz to/from 72Hz, have been very successful (they look quite good), although computationally expensive. However, 24 Hz to/from 60 Hz has proven to be quite difficult using motion analysis. At 24 φs, frames differ substantially, especially in having differing amounts of motion blur on each frame (as with the cockpit scenes from the movie "Top Gun"). This makes motion analysis, as well as subsequent frame rate conversion, difficult from a 24 φs source. Further, it is not possible to remove motion blur, so that even if motion analysis were possible on high-motion 24 φs scenes, the images would still be blurry (although they would move more smoothly with less stutter). Since motion analysis involves matching portions of an image, frames which have widely differing amounts of motion blur from adjacent frames become nearly impossible to match up. Thus, 24 φs source material from film (or electronic cameras) is a poor starting point for frame rate conversions to 50 Hz or 60 Hz video.
This leads to the conclusion that high frame rate electronic cameras are a much better image source than 24 φs electronic cameras. However, given the difficulties in converting from 60 φs video back to 24 φs film, 72 φs is a much better camera frame rate for eventual 24 φs compatibility.
Experiments have shown that a good quality 24 φs moving image can be derived from 72 φs frames through use of a very simple weighted frame filter. The best weightings for three consecutive frames (previous, current, and next) from a 72 φs source to yield one frame at 24 φs is centered about weightings of [0.1667, 0.6666, 0.1667]. However, any set of three frame weightings in the range [0.1, 0.8, 0.1] to [0.25, 0.5, 0.25] seem to work well. There is emphasis on the center frame, which helps strike a balance between the clarity of a single frame, due to the short motion blur, plus the needed blur from the adjacent frames in order to help smooth the stutter of 24 φs motion (by simulating 24 φs motion blur).
This weighting technique works well in about 95% of all cases, allowing this simple weighting function to provide the majority ofthe 24 φs conversions. For the remaining 5% or so ofthe cases, motion compensation can be used, as taught in U.S. patent application Serial No. 09/435,277. By having reduced the workload on the conversion process by a factor of 20 by this simple weighting technique, the remaining motion-compensated conversions become more practical when needed. It should also be noted that a 120 φs source can be used with five weightings to achieve a similar results at 24 φs. For example, weightings of [0.1, 0.2, 0.4, 0.2, 0.1] may be used. Also, 60 φs can be derived from 120 φs by taking every other frame, although the shorter open shutter duration will be noticeable on fast motion. In order to reduce this problem, an overlapping filter can also be used (e.g., preferably about [0.1667, 0.6666, 0.1667], but may be in the range [0.1, 0.8, 0.1] to [0.25, 0.5, 0.25]), repeating the low-amplitude weighted frames. Of course, even higher frame rates allow even more careful shaping ofthe temporal sample for deriving 24 φs and other frame rates. As the frame rates become very high, the techniques of U.S. Patent No. 5,465,119 and 5,737,027, assigned to the assignee ofthe present invention, begin to apply, since methods are needed to reduce the data rate within each frame in order to keep the data transfer rate manageable. However, on-chip parallel processing in the sensor (e.g., active pixel or CCD) can provide an alternative means to reduce the off- chip I O rates required. Given that 24 φs is desired for economic viability of new 72 φs (and other) frame rate formats, it is also important to be able to monitor the images at 24 φs, using the temporal filter weighting function described here (e.g., 0.1667, 0.6666, 0.1667). By doing so, the "blocking" (setting up) ofthe shots in a scene can be checked to ensure that the 24 φs results will look good (in addition to the 72 φs or other higher rate full-rate versions). In this way, the benefits of high frame rate capture are fully integrated with the capability to provide 24 φs international film and video release.
Thus, certain select higher frame rates form the most suitable basis for creating both a high-frame-rate electronic image source for the future, as well as being backward-compatible with today's existing 24 φs film and worldwide video releasing infrastructure.
Modular Bit Rate
It is useful in many video compression applications to "modularize" the bit rate. Variable bit rate systems have used continuously varying bit rates to attempt to apply more bits to faster changing shots. This can be done in a coarse way by giving each useful unit a different bit rate. Examples of suitable units include a range of frames (a "Group of Pictures," or GOP) or each P frame. Thus, for example, the bit rate might be constant within a GOP. However, for GOPs where high compression stress is detected (e.g., due to high motion or scene change), a higher constant bit rate 5 can be utilized. This is similar to the above-described layering technique of applying all ofthe bits in an enhancement layer to the base layer during periods of high stress (typically resetting at the next I frame). Thus, in addition to the concept of applying more bits to the base layer, more bits can be applied to single layer compressions, or to the base and enhancement layer (in the case of layered compression), so as to yield o high quality during periods of high stress.
It is typically the case that low bit rates can handle 90% ofthe time in a movie or live event. For the remaining 10% ofthe time, using 50% or 100% more bits generally will yield a near perfect encoding, while only increasing the overall bit count by 5% to 10%. This proves to be a very efficient way to get essentially visually perfect 5 encodings, while generally coding to a constant bit rate (thereby retaining most ofthe modularity and processing advantages of a constant bit rate).
The use of such higher bit rate periods can be either manually or automatically controlled. Automatic control is possible using the rate-control quantization scale factor parameter, which gets large (to keep the bit rate from greatly increasing) under 0 periods of high stress. Such high stress thus can be detected, signaling that either the remainder ofthe GOP should be coded at a higher bit rate, or else the GOP should be re-coded beginning at the starting I frame, using a higher bit rate. Using visual inspection, a manual selection can also be used to flag GOPs requiring a higher bit rate. 5 It is beneficial to real-time decoding to take advantage ofthe fact that GOP's generally have a specific size. Thus, using simple multiples of a GOP (e.g., a 50% or 100% increase in the number of bits for GOPs having high stress) also retains much of this advantage. FIG. 17 is a diagram of one example of applying higher bit rates to modular portions of a compressed data stream. Groups of pictures containing normal 0 scenes 1800, 1802 are allocated bits at a constant rate. When a GOP 1804 occurs that . contains a scene exhibiting a high level of stress (i.e., changes that are difficult for the compression process to compress as much as "normal" scenes), a higher number of bits (e.g., 50-100% additional) are allocated to that GOP to allow more accurate encoding ofthe scene.
It should be noted that many MPEG-2 implementations use a constant bit rate. Constant bit rate provides a good match with constant bit rate transport and storage media. Transport systems such as broadcast channels, satellite channels, cables, and fibers, all have a fixed constant total capacity. Also, digital compressed video tape storage systems have a constant tape playback rate, thereby yielding a constant recording or playback bit rate. Other MPEG-2 implementations, such as DirecTV/DSS, and DVD, use some form of variable bit rate allocation. In the case of DirecTV/DSS, the variability is a combination of scene stress in the current program vs. scene stress in adjacent TV programs which share a common multiplex. The multiplex corresponds to a tuned satellite channel and transponder, which has a fixed total bit rate. In the case of consumer video DVD, the digital optical disk capacity is 2.5 Gbytes, requiring that the MPEG-2 bit rate average 4.5 mbits/s for a two-hour movie. However, the optical disk has a peak reading rate capability of 100% higher, at 9 mbits/s. For a shorter movie, the average rate can be higher, up to the full 9mbits/s. For a two-hour movie, the way that the bit rate achieves an average of 4.5mbits/s is that a rate above this is used for scenes having high scene stress (high change due to rapid scene motion), while a rate below this average is used during low scene stress (low change due to little motion).
The bit rate in MPEG-2 and MPEG-4 is held constant by a combination of modeling of a virtual decoder buffer's capacity, and by varying the quantization parameter to throttle the bit rate emitted from the encoder. Alternatively, a constant quantization parameter will yield a variable number of bits, in proportion to scene change and detail, also known as scene "entropy". A constant quantization parameter yields relatively constant quality, but variable bit rate. A varying quantization parameter can be used in conjunction with a size bounded decoder buffer to smooth out any variability and provide a constant bit rate. The sharing of many channels in a multiplex is one method that can support variable bit rate, as with DirecTV, or with standard definition signals in the ACATS/ATSC 19.3 mbits/s 6Mhz multiplex. The statistics of high entropy shows (fast sports, like hockey) paired with low entropy shows (like talk shows), allows instantaneous tradeoff in the application of bits to shows having more entropy. Slow periods in one show use fewer bits, providing more bits for a fast moving simultaneous alternate show in the same multiplex.
It should be noted that these variable bit rate systems have a peak bit rate, usually somewhere near 100% above the average. Thus, these systems become constant bit rate systems at the highest bit rate, limiting the peak bit rate available for periods of continued high scene stress. There is also a limit to the input bit rate in some MPEG-2 decoder systems, also limiting the peak bit rate in such variable bit rate systems. However, this limit on peak input bit rate is gradually rising well above these other limits, as decoders improve.
The general concept of each of these prior bit rate control systems is that there is a small memory buffer in the decoder, holding somewhere between a fraction of a frame and a few frames of moving image. At the time this decoder bit rate buffer was conceived, around 1990, there was concern that the memory cost of this buffer in decoders would have a significant affect on the decoder's price. However, as ofthe present, the cost of this buffer has proven insignificant. In fact, many seconds' worth of buffer is now an insignificant cost. It may be extrapolated that, in the near future, the bit receiving memory buffer may hold many minutes of video information at insignificant cost. Further, the cost of disk and other storage media has also fallen rapidly, while capacity has increased rapidly. It is therefore also reasonable to spool the compressed bitstream to disk or other storage memory systems, thereby yielding many hours or days worth of storage capacity. This is currently being done by commercially available harddrive based home video recorders.
One fundamental issue remains, however, which is that there will be time delay while bits wait in a compressed bit buffer. For broadcast television and movie distribution, a delay of several seconds or tens of seconds would have little affect on viewing, as long as an auxiliary selection stream is available to guide ongoing program "tune-in" or "movie selection" or where the initial start (of a movie, for example) uses a shortened delay through a small initial buffer. However, for teleconferencing or live interactive events, a small fast running buffer may be required in order to minimize delay. With the exception of live interactive and teleconferencing applications, inexpensive large buffers can be utilized to improve quality. In light of these trends, the architecture of variable and constant bit rate compression techniques can be significantly improved. These improvements include:
• Greatly increasing the buffer size in the decoder buffer model, thereby providing much ofthe benefits of variable bit rate and constant bit rate simultaneously. • Pre-loading of "interstitial" show titles, to support instantaneous change to the titles, while decoder buffers begin to fill.
• Utilizing a partially-filled FIFO (first in, first out) decoder bit rate buffer at the beginning of newly starting programs or movies and gradually increasing the buffer fullness (and therefore delay) as the program progresses after starting. • Pre-loading into decoder bit memory (e.g. , using a second FIFO, main memory, or even spooling to disk) increased bit rate "modules" (using the modular bit rate concept described above) to augment the average bit rate during periods of high scene stress. Such "pre-loading" can allow periods of bit rate which exceed the average bit rate in constant bit rate channels, but also which exceed the maximum bit rate in variable bit rate systems.
• In the layered structure ofthe invention, all ofthe bits in the average (or constant) bit rate stream could be shunted to the base layer during scenes having high scene stress. However, the enhancement layer bits for a scene can be pre-loaded for that scene, and can also be played out using a timing marker for synchronization. Note again that maximum (or constant) bit rate limits in transport and/or playback can be exceeded for periods of time (limited only by amount of available buffer space) using this technique. Multi-Layer DCT Structure
Variable DCT Block Size
Fundamental to a layered DCT structure is the harmonic alignment ofthe transform wavelengths. For example, FIG. 18 graphically illustrates the relationships of DCT harmonics between two resolution layers. In a currently optimal two layer configuration ofthe invention, the base layer utilizes DCT coefficients using an arithmetic harmonic series having frequencies of 1, 2, 3, 4, 5, 6, and 7 times the 8x8 pixel DCT block size 1900. At the factor-of-two resolution enhancement layer, these base layer harmonics then map to frequencies of 1/2, 1, 3/2, 2, 5/2, 3, and 7/2 ofthe corresponding enhancement layer DCT block 1902. Although there is no penalty for the 112 term, since its frequency is entirely held in the base layer, the remaining terms only partially harmonize with the enhancement layer. For example, frequencies of 2, 4, and 6 times the macroblock size from the base layer are aligned with frequencies of 1, 2, and 3 times the macroblock size from the enhancement layer. These terms form a natural signal-to-noise ratio (SNR) layering, as if additional precision were applied to these coefficients in the base layer. The 3, 5, and 7 terms from the base layer are non- harmonic with the enhancement layer, and therefore represent orthogonality to the base layer only, providing no synergy with the enhancement layer. The remaining terms in the enhancement layer, 4, 5, 6, and 7, represent additional detail which the enhancement layer can provide to the image, without overlap with the base layer. FIG. 19 graphically illustrates the similar relationships of DCT harmonics between three resolution layers, showing a highest enhancement layer 1904.
It can be seen that there is only partial orthogonality and partial alignment in this structure. While this alignment and orthogonality is mostly beneficial, the phase alignment ofthe DCT coding series was never optimized for two (or more) spatial resolution layers. Rather, the DCT was designed as a single set of orthogonal basis functions utilizing phase characteristics which eliminated the phase-carrying imaginary terms from the Fourier transform series. While the DCT is demonstrably adequate in coding performance in a two-layer spatial coding structure, these issues of layer orthogonality and phase relationships become central to the expansion ofthe layered structure to three or four spatial resolution layers.
A solution to providing cross-layer orthogonality is to utilize different DCT block sizes for each resolution layer. For example, if a given layer doubles the resolution, then the DCT block size will be twice as large. This results in a harmonically aligned resolution layering structure, providing optimal coding efficiency due to optimal inter-layer coefficient orthogonality.
FIG. 20 is a diagram showing various DCT block sizes for different resolution layers. For example, a 4x4 pixel DCT block 2000 could be used at the base layer, an 8x8 pixel DCT block 2002 could be utilized at the next layer up, a 16x16 pixel DCT block 2004 could be utilized at the third layer, and a 32x32 pixel DCT block 2006 could be utilized at the fourth layer. In this way, each layer adds additional harmonic terms in full orthogonality to the layer(s) below. Optionally, additional precision (in the SNR sense) can be added to previously covered coefficient terms. For example, the 16x16 pixel subset 2008 in the 32x32 pixel block 2006 can be used to augment (in an SNR improvement sense) the precision ofthe 16x16 pixel DCT block 2004.
Motion Vectors
In MPEG-2, macroblocks corresponding to motion vectors consist of 16x16 pixels, organized as four 8x8 DCT blocks. In MPEG-4, each macroblock can optionally be further subdivided into 8x8 regions, corresponding to the DCT blocks, each having their own motion vector.
Even though the DCT blocks preferably have differing sizes in each layer, the motion compensation macroblocks need not be constrained by this structure. The simplest structure is where the single motion vector for each base layer motion compensation macroblock applies to all higher layers as well, eliminating motion vectors from all enhancement layers entirely, since the motion is specified by the base layer's motion vector for all layers together. A more efficient structure, however, is to allow each layer to independently select (1) no motion vector (i.e., use the base layer motion vector), (2) additional sub-pixel precision for the base layer's motion vector, or (3) split each motion compensation macroblock into two, four, or other numbers of blocks each having independent motion vectors. The technique of overlapped block motion compensation (OBMC) in MPEG-4 could be utilized to smooth the transition between the independent blocks motion compensation being moved. The use of negative-lobed filters for sub-pixel placement, as specified in other parts of this description, is also beneficial to the motion compensation of this DCT layer structure. Thus, each DCT block at each layer may be split into as many motion vector blocks for motion compensation as are optimal for that layer. FIG. 21 is a diagram showing examples of splitting of motion compensation macroblocks for determining independent motion vectors. For example, the base layer, if constructed using 4x4 pixel DCT blocks 2100, could utilize from one (shown) to as many as 16 motion vectors (one for each pixel), or even utilize sub-pixel motion vectors. Correspondingly, each higher level can split its larger corresponding DCT block 2102, 2104, 2106 as appropriate, yielding an optimal balance between coding prediction quality (thus saving DCT coefficient bits) vs. the bits required to specify the motion vectors. The block split for motion compensation is a tradeoff between the bits used to code the motion vectors and the improvement in picture prediction.
The use of guide vectors from motion vectors for lower layers to predict each higher layer's motion vectors, as described in other portions of this description, also improves coding efficiency and effectiveness.
Variable Length Coding Optimization
The variable length codes (such as Huffman or arithmetic codes) used by MPEG-1, MPEG-2, MPEG-4, H.263, and other compression systems (including wavelets and other DCT and non-DCT systems) are selected based upon demonstrated efficiency on a small group of test sequences. These test sequences are limited in the types of images, and only represent a relatively narrow range of bit rate, resolution, and frame rate. Further, the variable length codes are selected based upon average performance over each test sequence, and over the test sequences as a group.
Experimentation has shown that a substantially more optimal variable length coding system can be obtained by (1) applying specific variable length coding tables to each frame and (2) selecting the most optimal codes for that particular frame. Such a selection of optimal variable length codes can applied in units smaller than a frame (a part or region of a frame), or in groups of several frames. The variable length codes used for motion vectors, DCT coefficients, macroblock types, etc., can then each be independently optimized for the instantaneous conditions of a given unit (i.e., frame, sub-frame, or group of frames) at that unit's current resolution and bit rate. This technique is also applicable to the spatial resolution enhancement layers described in other parts of this description.
The selection of which group of variable length codes is to be used can be conveyed with each frame (or subpart or group) using a small number of bits. Further, custom coding tables can be downloaded where reliable data transmission or playback is available (such as with optical data disk or optical fiber networks).
Note that the existing coding tables used by MPEG-1, MPEG-2, MPEG-4, H.263, DVC-Pro/DV, and other compression systems are pre-defined and static. Thus, application ofthe this aspect ofthe invention would not be backwards compatible, but may be forward compatible with future coding systems.
Augmentation System For MPEG-2 and MPEG-4
At present, there is a large installed base of MPEG-2 capable decoders. For example both DVD players and DirecTV satellite receivers are now in millions of homes. The improvement which MPEG-4 video compression coding could offer beyond MPEG-2 is not yet available, since MPEG-4 is incompatible with MPEG-2. However, MPEG-4 and MPEG-2 are both motion-compensated DCT compression system, sharing a common basic structure. The composition system in MPEG-4's video coding system is fundamentally different from MPEG-2, as are some other expanded features. In this discussion, only the full frame video coding aspects of MPEG-4 are being considered.
Although there are many differences between MPEG-4 and MPEG-2, the following are the main differences:
(1) MPEG-4 can optionally split a 16x16 macroblock into four 8x8 blocks, one for each DCT, each having an independent motion vector. (2) MPEG-4 B-frames have a "direct" mode, which is a type of prediction. (3) MPEG-4 B-frames do not support "I" macroblocks, unlike MPEG-2 which does support "I" macroblocks in B-frames.
(4) The DCT coefficients in MPEG-4 can be coded by more elaborate patterns than with MPEG-2, although the well-known zigzag pattern is common to both MPEG-2 and MPEG-4.
(5) MPEG-4 supports 10-bit and 12-bit pixel depths, whereas MPEG-2 is limited to 8 bits.
(6) MPEG-4 supports quarter-pixel motion vector precision, whereas MPEG-2 is limited to half-pixel precision.
Some of these differences, such as the B-frame "direct" mode and "I" macroblocks, are fundamental incompatibilities. However, both of these coding modes are optional, and an encoder could chose to use neither of them (at a small efficiency loss), thereby eliminating this incompatibility. Similarly, an encoder could restrict the coding patterns in MPEG-4 for DCT coefficients to provide for better MPEG-2 commonality (again at a small efficiency loss).
The thee remaining major items, the 8x8 four-way block split, the quarter- pixel motion vector precision, and the 10-bit and 12-bit pixel depths, could be considered to be "augmentations" to the basic structure which MPEG-2 already provides. This aspect ofthe invention takes advantage ofthe fact that these
"augmentations" can be provided as separate constructs. Accordingly, they can be coded separately and conveyed as a separate augmentation stream together with a standard MPEG-2 or MEPG-4 stream. This technique can also be used with MPEG-1, H.263, or any other video coding system which shares a common motion- compensated DCT structure. FIG. 22 is a block diagram showing an augmentation system for MPEG-2 type systems. A main compressed data stream 2200 (shown as including motion vectors, DCT coefficients, macroblock mode bits, and I, B, and P frames) is conveyed to a conventional MPEG-2 type decoder 2202 and to a parallel enhanced decoder 2204. Concurrently, an enhanced data stream 2206 (shown as including quarter-pixel motion vector precision, 8x8 four-way block split motion vectors, and 10-bit and 12-bit pixel depths) is conveyed to the enhanced decoder 2204. The enhanced decoder 2204 would combine the two data streams 2200, 2206 and decode them to provide an enhanced video output. Using this structure, any coding enhancements can be added to any motion-compensated DCT compression system. The use of this structure can be biased by an encoder toward more optimal
MPEG-2 decoding, or toward more optimal enhanced decoding. The expectation is that such enhanced decoding, by adding MPEG-4 video coding improvements, would be favored, to achieve the optimal enhanced picture quality, with a small compromise in quality to the MPEG-2 decoded picture. For example, in the case of MPEG-4 enhancements to MPEG-2 video coding, the MPEG-2 motion vectors can be used as "predictors" for the four-way split motion vectors (in those cases where MPEG-4 chooses to split four ways), or may be used directly for non-split 16x16 macroblocks. The quarter pixel motion vector resolution can be coded as one additional bit of precision (vertically and horizontally) in the enhanced data stream 2206. The extra pixel depth can be coded as extra precision to the DCT coefficients prior to applying the inverse DCT function.
The spatial resolution layering which is a principal subject of this invention performs most optimally when the base layer is as perfectly coded as possible. MPEG- 2 is an imperfect coding, yielding degraded performance for resolution enhancement layers. By using this augmentation system, the base layer can be improved, for example, by using the MPEG-4 improvements described above (as well as other improvements set forth in this description) to augment the MPEG-2 data stream that encodes the base layer. The resulting base layer, with accompanying enhancement data stream, will then have most ofthe quality and efficiency that would have been obtained using an improved base layer which would have resulted from better coding (such as with MPEG-4 and the other improvements of this invention). The resulting improved base layer can then have one or more resolution enhancement layers applied, using other aspects of this invention.
The other improvements of this invention, such as the better filters with negative lobes for motion compensation, can also be invoked by the augmented enhanced decoder, thus yielding further improvements beyond those provided by MPEG-4 or other motion compensated compression systems.
Guide Vectors For The Spatial Enhancement Layer
Motion vectors comprise a large portion ofthe allocated bits within each resolution enhancement layer created in accordance with the invention. It has been determined that it is possible to substantially reduce the number of bits required for enhancement layer motion vectors by using the corresponding motion vectors at the same position in the base layer as "guide vectors". The enhancement layer motion vectors are therefore coded by only searching for a small search range about the corresponding guide vector center from the base layer. This is especially important with MPEG-4 enhancement layers, since each macroblock can optionally have 4 motion vectors, and since quarter-pixel resolution of motion vectors is available.
FIG. 23 is a diagram showing use of motion vectors from a base layer 2300 as guide vectors for a resolution enhancement layer 2302. A motion vector 2304 from the base layer 2300, after expansion up to scale ofthe resolution enhancement layer 2302, serves as a guide vector 2304' for refinement ofthe motion vectors for the enhancement layer 2302. Accordingly, only a small range need be searched to find the corresponding enhancement layer 2302 motion vector 2306. The process is the same for all ofthe motion vectors from the base layer. For example, in MPEG-4 a 16x16 pixel base layer macroblock may optionally be split into four 8x8 pixel motion vector blocks. A corresponding factor-of-two enhancement layer would then utilize the co- located motion vectors from the base layer as guide vectors. In this example, a motion vector from one ofthe 8x8 motion vector blocks in the base layer would guide the search for a motion vector in a corresponding 16x16 pixel macroblock in the enhancement layer. This 16x16 block could optionally be further split into four 8x8 motion vector blocks, all using the same corresponding base layer motion vector as a guide vector.
These small search range motion vectors in the enhancement layer are then coded much more efficiently (i.e., fewer bits are required to code the smaller enhancement layer motion vectors 2306). This guide-vector technique is applicable to MPEG-2, MPEG-4, or other appropriate motion-compensated spatial resolution enhancement layer(s).
Enhancement Modes
FIGS. 24A-24E are data flow diagrams showing on example professional level enhancement mode. This figures shows picture data (including intermediate stages) in the left column, processing steps in the middle column, and ouφut in the right column. It should be noted that this is just one example of how to combine a number ofthe processing steps described herein. Different combinations, simpler as well as more complex, can be configured to achieve different levels of compression, aspect ratios, and image quality.
FIG. 24A shows an initial picture 2400 at 2kxlk pixels. Down filter 2402 this image to lkx512 pixels 2404. Create motion vectors 2406 from the initial picture and output as a file 2407. Compress/decompress 2408 the lkx512 pixel image 2404 to a lkx512 decompressed image 2410 and output the compressed version as the base layer 2412, along with the associated motion vector files 2416. Expand 2418 the lkx512 decompressed image 2410 as a 2kxlk image 2420. Expand 2422 the lkx512 image 2404 as a 2kxlk image 2424. Subtract 2426 the 2kxlk image 2420 from the original image 2400 to create a 2kxlk difference picture 2428.
Subtract 2430 the 2kxlk image 2424 from the original image 2400 to create a 2kxlk difference picture 2432. Reduce 2434 the amplitude ofthe 2kxlk difference picture 2432 to a selected amount (e.g., 0.25) to create a 2kxlk scaled difference picture 2436. Add 2438 the 2kxlk scaled difference picture 2436 to the 2kxlk difference picture 2428 to create a 2kxlk combined difference picture 2440.
Encode/decode 2442 the combined difference picture 2440 using the original motion vectors and ouφut an encoded enhancement layer 2444 (MPEG-2, in this example), and a 2kxlk decoded enhanced layer 2246. Add 2448 the 2kxlk decoded enhanced layer 2246 to the 2kxlk image 2420 to create a 2kxlk reconstructed full base plus enhancement image 2450. Subtract 2452 the original image 2400 from the 2kxlk reconstructed full base plus enhancement image 2450 to create a 2kxlk second layer difference picture 2454. Increase 2456 the amplitude ofthe 2kxlk second layer difference picture 2454 to create a 2kxlk difference picture 2458. Then extract the red channel information 2458, the green channel information 2460, and the blue channel information 2462 to create respective red difference 2464, green difference 2466, and blue difference 2468 images. Using the motion vector file 2407: encode/decode 2470 a second red layer from the red difference picture 2464 as a red second enhancement layer 2472, and a decoded red difference image 2474; encode/decode 2476 a second green layer from the green difference picture 2466 as a green second enhancement layer 2478, and a decoded green difference image 2480; and encode/decode 2482 a second blue layer from the blue difference picture 2468 as a blue second enhancement layer 2484, and a decoded blue difference image 2486. Combine 2488 the decoded red difference image 2474, the decoded green difference image 2480, and the decoded blue difference image 2486 into a decoded RGB difference image 2490. Decrease 2492 the amplitude ofthe decoded RGB difference image 2490 to create a second decoded RGB difference image 2494. Add 2496 the second decoded RGB difference image 2494 to the 2kxlk reconstructed full base plus enhancement image 2450 to create a 2kxlk reconstructed second enhancement layer image 2498. Subtract 2500 the 2kxlk reconstructed second enhancement layer image 2498 from the original image 2400 to create a 2kxlk final residual image 2502. This 2kxlk final residual image 2502 is then losslessly compressed 2504 to create separate red, green, and blue final difference residuals 2506.
COMPUTER IMPLEMENTATION
The invention may be implemented in hardware or software, or a combination of both. However, preferably, the invention is implemented in computer programs executing on one or more programmable computers each comprising at least a processor, a data storage system (including volatile and non- volatile memory and/or storage elements), an input device, and an ouφut device. Program code is applied to input data to perform the functions described herein and generate ouφut information. The ouφut information is applied to one or more ouφut devices, in known fashion.
Each such program may be implemented in any desired computer language (including machine, assembly, or high level procedural, logical, or object oriented programming languages) to communicate with a computer system. In any case, the language may be a compiled or interpreted language. Each such computer program is preferably stored on a storage media or device
(e.g., ROM, CD-ROM, or magnetic or optical media) readable by a general or special purpose programmable computer system, for configuring and operating the computer when the storage media or device is read by the computer system to perform the procedures described herein. The inventive system may also be considered to be implemented as a computer-readable storage medium, configured with a computer program, where the storage medium so configured causes a computer system to operate in a specific and predefined manner to perform the functions described herein.
CONCLUSION
Different aspects ofthe invention that are considered to be novel include (without limitation) the following concepts:
• Use of 72 φs as a source frame rate for electronic cameras, in order to provide compatibility with the existing worldwide 24 φs film and video infrastructure, while allowing the benefits of high frame rate for new electronic video systems.
• Conversion to 60 φs from 72 φs and/or 120 φs using the motion compensation and frame rate conversion techniques from U.S. patent application Serial No. 09/435,277 (entitled "System And Method For Motion Compensation and Frame Rate Conversion", filed 11/5/1999).
• Conversion to 24 φs from 72 φs using filters with weightings in the range [0.1, 0.8, 0.1] to [0.25, 0.5, 0.25], and conversion to 24 φs from 120 φs using weightings of approximately [0.1, 0.2, 0.4, 0.2, 0.1].
• Conversion to 60 φs from 120 φs using overlapping sets of three frames (advanced two 120ths for each one 60th frame) using weightings in the range [0.1,
0.8, 0.1] to [0.25, 0.5, 0.25].
• Using motion compensation and frame techniques from U.S. patent application Serial No. 09/435,277 (entitled "System And Method For Motion Compensation and Frame Rate Conversion", filed 11/5/1999) to increase the motion blur and convert the frame rate from 72 φs (or other higher rate) source to 24 φs, on the small percentage of scenes where the generally preferred simple weightings may be less than the desired quality.
• Using 24 φs monitoring, via the weighting functions described above, while shooting using a higher (72 φs, 120 φs, etc.) frame rate. • Simultaneous release ofthe derived 24 φs result together with the original higher frame rate.
• De-graining and/or noise-reducing filtering prior to layered encoding.
• Re-graining or re-noising after decoding, as a creative effect.
• De-interlacing prior to layered compression. Applying a three-field-frame de-interlacer prior to either single and multi-layer compression.
Upfiltering a picture prior to either single and multi-layer compression, thereby providing improved color resolution.
Adjusting the size of a sub-region within an enhancement layer, and the relative proportion ofthe bits allocated to the base and enhancement layer.
Treating vertical and horizontal relationships as independent, such that the fractional relationships can be independent and different.
Allowing higher bit rates for compression units (such as the GOP) during periods of high compression stress (either automatically, by detecting high values of rate control quantization parameter, or manually controlled).
Using "modularized" bit rates wherein natural units of compression and layered compression systems can utilize increased bit rates in modular units.
Pre-loading a decompression buffer(s) with modular units of increased bit rate for use with compression or layered compression systems.
Using constant bit rate systems with one or more layers ofthe present layered compression system.
Using variable bit rate systems with one or more layers ofthe present layered compression system. • Using combined fixed and variable bit rate systems used with various layers ofthe present layered compression system.
Using correspondingly larger DCT block size and additional DCT coefficients for use in resolution layering (also called "spatial scalability"). For example, if a given layer doubles the resolution, then the DCT block size will be twice as large. This results in a harmonically aligned resolution layering structure, providing optimal coding efficiency due to optimal inter-layer coefficient orthogonality. • Using multiple motion vectors per DCT block, so that both large and small DCT blocks can optimize the tradeoff between motion vector bits and improved motion compensated prediction. • Using negative-lobed upsizing and downsizing filters, particularly truncated sine filters. • Using negative-lobed motion compensation displacement filters.
• Selection of optimal variable length codes on a relatively instantaneous basis, such as each frame, each region of a frame (such as several scanlines or macroblock lines or each quadrant), or every several frames. • Using an augmentation stream to add improved coding features to existing compression systems, thereby providing backward compatibility as well as improved quality using a new enhanced decoder.
• Using an enhanced decoded picture to provide a higher quality base layer for resolution layering. • Sharing of coding elements between similar moving image coding systems to provide backward compatibility as well as a path to improvement.
• Consideration in the encoding process of generating compressed bitstreams partially common to two types of decoders, including provision for favoring one or the other. • Using base layer motion vectors as guide vectors to center the range of motion vectors used in the enhancement layer.
• Application of combinations ofthe above techniques to enhancement layers, or to improve MPEG-1, MPEG-2, MPEG-4, H.263, DNC-Pro/DN, and other compression systems, including wavelet-based systems.
A number of embodiments ofthe invention have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. For example, while the preferred embodiment uses MPEG-2 or MPEG-4 coding and decoding, the invention will work with any comparable standard that provides equivalents of I, P, and/or B frames and layers. Accordingly, it is to be understood that the invention is not to be limited by the specific illustrated embodiment, but only by the scope ofthe appended claims.

Claims

WHAT IS CLAIMED IS:
1. A method for creating an enhancement layer for a base layer of an image encoding system, including: upfiltering and expanding the base layer to an expanded base region; creating an additional area region surrounding the expanded base region by padding the expanded base region with a uniform mid-gray pixel value; and creating an enhancement layer that provides additional picture information, the enhancement layer including a difference picture having a small range of possible pixel values for an area that coincides with the expanded base region, and a large range of pixel values for an area that coincides with the additional area region.
2. The method of claim 1, further including encoding the enhancement layer as part of a picture stream that includes the base layer.
3. The method of claim 2, further including decoding the enhancement layer.
4. The method of claim 1 , wherein the difference picture includes motion vectors, and further including constraining the motion vectors not to point into the additional area region.
5. The method of claim 4, further including determining the motion vectors based on macroblocks, the macroblocks being aligned such that no macroblock spans the boundary between the expanded base region and the additional area region surrounding the expanded base region.
6. The method of claim 1 , wherein the base layer and the enhancement layer have a resolution ratio selected from one of 3/2, 4/3, and exact factors of 2.
7. The method of claim 1 , wherein the difference picture is centered in the enhancement layer.
8. The method of claim 1 , further including continuously repositioning the difference picture with respect to the enhancement layer from image to image.
9. A method for creating a lower resolution image from a higher resolution image in an image encoding system, including applying a downsizing filter to the higher resolution original image, the downsizing filter having a positive central lobe, two negative lobes each adjacent opposite sides ofthe positive central lobe, and small positive lobes corresponding to and adjacent each negative lobe, each small positive lobe separated from the positive central lobe by a corresponding negative lobe.
10. The method of claim 9, wherein the extent of the downsizing filter is limited to the small positive lobes.
11. The method of claim 9, wherein the relative amplitudes of the positive central lobe, the negative lobes, and the small positive lobes are approximated by a truncated sine function.
12. The method of claim 9, wherein the relative amplitude of the positive central lobe is approximated by a truncated sine function, and the relative amplitudes ofthe small positive lobes and the negative lobes are approximated as 1/2 to 2/3 of a truncated sine function.
13. A method for creating an enlarged image from a decompressed base or enhancement image layer in an image encoding system, including applying a pair of upsizing filters to the decompressed base or enhancement image layer, each upsizing filter having a positive central lobe, and two negative lobes each adjacent opposite sides ofthe positive central lobe, wherein the peaks ofthe positive central lobe of each upsizing filter are asymmetrically spaced with respect to each other.
14. The method of claim 13 , wherein the extent of the upsizing filter is limited to the negative lobes.
15. The method of claim 13 , wherein the relative amplitudes of the positive central lobe is approximated by a truncated sine function, and the relative amplitude ofthe negative lobes is less than would be approximated by a truncated sine function.
16. The method of claim 13, wherein the relative amplitude ofthe positive central lobe is approximated by a truncated sine function, and the relative amplitudes ofthe negative lobes are approximated as 1/2 to 2/3 of a truncated sine function.
17. A method for creating an enhanced detail image from an original uncompressed base layer input image created from an original high resolution image in an image encoding system, including: applying a Gaussian upsizing filter to the original uncompressed base layer image to create an expanded image; creating a difference image by subtracting the expanded image from the original high resolution image; and multiplying the difference image by a weight factor.
18. The method of claim 17, wherein the weight factor is in the range of approximately 4% to approximately 35%.
19. The method of claim 17, wherein the encoding system conforms to the MPEG-4 standard and the weight factor is in the range of approximately 4% to approximately 8%.
20. The method of claim 17, wherein the encoding system conforms to the MPEG-2 standard and the weight factor is in the range of approximately 10% to approximately 35%.
21. A method for enhancing image quality in an image encoding system, including: applying at least one of a de-graining filter or a noise-reducing filter to an original digital image to create a first processed image; and encoding the first processed image in the image encoding system into a compressed image.
22. The method of claim 21, wherein the original image comprises separate color channel images having uncorrelated noise characteristics, further including applying a separate noise-reducing filter to each one of such separate color channel images.
23. The method of claim 21 , further including: decoding the compressed image into a decompressed image; and applying at least one of a re-graining filter or a re-noising filter to the decompressed image.
24. A method for enhancing image quality in an image encoding system, including: applying a field de-interlacer to each of a series of image fields to create a corresponding series of field-frames; applying a field-frame de-interlacer to a series of at least three sequential field-frames to create a corresponding series of de-interlaced image frames; and encoding the series of de-interlaced image frames in the image encoding system into a series of compressed images.
25. The method of claim 24,wherein each image field comprises lines, and applying the field de-interlacer includes: replicating each line of an image field; and synthesizing, for each adjacent pair of lines ofthe image field, a line intermediate such pair of lines by averaging such pair of lines.
26. The method of claim 24,wherein applying the field-frame de-interlacer includes synthesizing a de-interlaced image frame, for each of a previous field-frame, a current field-frame, and a next field-frame, as a weighted average of such field- frames.
5 27. The method of claim 26, wherein the weights for the previous field- frame, current field-frame, and next field-frame are approximately 25%, 50%, and 25%, respectively.
28. The method of claim 24, wherein each de-interlaced image frame and each field-frame comprise pixel values, and further including: o comparing the difference between each corresponding pixel value of each de-interlaced image frame and each corresponding current field-frame to a threshold value to generate a difference value; and selecting, for each final pixel value for the de-interlaced image frame, a corresponding pixel value from the current field-frame if the difference value is within 5 a first threshold comparison range, and a corresponding pixel value from the de- interlaced image frame if the difference value is within a second threshold comparison range.
29. The method of claim 24, wherein the threshold value is selected from the range of approximately 0.1 to approximately 0.3.
0 30. The method of claim 28, further including smooth-filtering each de- interlaced image frame and current field-frame before comparing.
31. The method of claim 30, wherein smooth-filtering includes down filtering followed by up filtering.
32. The method of claim 24, wherein each de-interlaced image frame and 5 each field-frame comprise pixel values, fiirther including adding a weighted amount of each current field-frame to a weighted amount of each de-interlaced image frame.
33. The method of claim 32, wherein the weighted amount of each current field-frame is 1/3 and the weighted amount of each de-interlaced image frame is 2/3.
34. A method for enhancing image quality of video images in an image encoding system, wherein the video images comprise digital pixel values that represent non-linear signals, the method including: converting the digital pixel values of each video image representing non-linear signals to a linear representation to create a linearized image; applying a transformation function to at least one linearized image to create a transformed image; and converting each transformed image back into a video image including digital pixel values that represent non-linear signals.
35. A method for encoding video images, including: downsizing the horizontal and vertical dimensions of an original image by respective first and second selected simple fraction factors to create a first intermediate image; encoding the first working image as a compressed base layer; decompressing the base layer and upsizing the result by the inverse of the selected simple fraction factors to create a second intermediate image; upsizing the first intermediate image by the inverse ofthe selected simple fraction factors and subtracting the result from the original image and weighting such result to create a first intermediate result; subtracting the second intermediate image from the original image to create a second intermediate result; adding the first intermediate result and the second intermediate result to create a third intermediate image; and encoding the third intermediate image to create an enhancement layer.
36. The method of claim 35, further including cropping and edge feathering the third intermediate image before encoding.
37. The method of claim 35, wherein the first and second simple fraction factors are each selected from one of 1/3, 1/2, 2/3, and 3/4.
38. A method for enhancing image quality in an image encoding system, including: applying a median filter to horizontal pixel values of a digital video image; applying a median filter to vertical pixel values ofthe digital video image; and averaging the results ofthe filtering ofthe horizontal pixels and vertical pixel values to create a noise-reduced digital video image.
39. The method of claim 38, further including: applying a median filter to diagonal pixel values ofthe digital video image; and averaging the results ofthe filtering ofthe diagonal pixel values with the noise-reduced digital video image.
40. A method for enhancing image quality in an image encoding system, including: applying a temporal median filter to corresponding pixel values of a previous digital video image, a current digital video image, and a next digital video image to create a noise-reduced digital video image.
41. The method of claim 40, further including: comparing the difference between each corresponding pixel value of each noise-reduced digital video image and each corresponding current digital video image to a threshold value to generate a difference value; and selecting, for each final pixel value for the noise-reduced digital video image, a corresponding pixel value from the current digital video image if the difference value is within a first threshold comparison range, and a corresponding pixel value from the noise-reduced digital video image if the difference value is within a second threshold comparison range.
42. The method of claim 41 , wherein the threshold value is selected from the range of approximately 0.1 to approximately 0.3.
43. A method for enhancing image quality in an image encoding system, including: applying a horizontal median filter to horizontal pixel values of a current digital video image; applying a vertical median filter to vertical pixel values ofthe current digital video image; applying a temporal median filter to corresponding pixel values of a previous digital video image, the current digital video image, and a next digital video image; and applying a median filter to corresponding pixel values produced by each ofthe horizontal, vertical, and temporal filters to create a noise-reduced digital video image.
44. A method for enhancing image quality in an image encoding system, including creating a noise-reduced digital video image comprising a linear weighted sum of five terms: (1) a current digital video image;
(2) an average of horizontal and vertical medians ofthe current digital video image;
(3) a thresholded temporal median;
(4) an average of horizontal and vertical medians ofthe thresholded temporal median; and
(5) a median ofthe thresholded temporal median and horizontal and vertical medians ofthe current digital video image.
45. The method of claim 44, wherein the weights of the five terms are approximately 50%, 15%, 10%, 10%, and 15%, respectively.
46. The method of claim 44, wherein the weights of the five terms are approximately 35%, 20%, 22.5%, 10%, and 12.5%, respectively.
47. The method of claim 44, further including: determining a motion vector for each nxn pixel region ofthe current digital video image with respect to at least one previous digital video image and at least one subsequent digital video image; applying a center weighted temporal filter to each nxn pixel region of the current digital video image and corresponding motion- vector offset nxn pixel regions ofthe at least one previous digital video image and at least one subsequent digital video image to create a motion-compensated image; and adding the motion-compensated image to the noise-reduced digital video image.
48. A method for enhancing image quality in an image encoding system, including: determining a motion vector for each nxn pixel region of a current digital video image with respect to at least one previous digital video image and at least one subsequent digital video image; and applying a center weighted temporal filter to each nxn pixel region of the current digital video image and corresponding motion- vector offset nxn pixel regions ofthe at least one previous digital video image and at least one subsequent digital video image to create a motion-compensated image.
49. The method of claim 48, wherein each digital video image is a de- interlaced field-frame.
50. The method of claim 48, wherein each digital video image is a three- field-frame de-interlaced image.
51. The method of claim 48, wherein each digital video image is a thresholded three-field-frame de-interlaced image.
52. The method of claim 48, wherein the center weighted temporal filter is a three-image temporal filter having weights for each of such images of approximately 25%, 50%, and 25%, respectively.
53. The method of claim 48, wherein the center weighted temporal filter is a five-image temporal filter having weights for each of such images of approximately 10%, 20%, 40%, 20%, and 10%, respectively.
54. A method for enhancing image quality in an image encoding system, including: applying a normal down filter to an image to create a first intermediate image; applying a Gaussian up filter to the first intermediate image to create a second intermediate image; and adding a weighted fraction of the second intermediate image to a selected image to create an image having reduced high frequency noise.
55. The method of claim 54, wherein the weighted fraction is between approximately 5% and 10% ofthe second intermediate image.
56. A method for enhancing image quality in an image encoding system, including: applying a down filter to a noise-filtered original resolution image to create a first intermediate image at a base layer resolution; applying a normal down filter to the first intermediate image to create a second intermediate image; applying a Gaussian up filter to the second intermediate image to create a third intermediate image; creating a noise-reduced digital video image comprising a linear weighted sum of three terms:
(1.) the first intermediate image;
(2.) an average of horizontal and vertical medians ofthe first intermediate image; and
(3.) the third intermediate image.
57. The method of claim 55, wherein the weights ofthe three terms are approximately 70%, 22.5%, and 7.5%, respectively.
58. A method for enhancing image quality in an image encoding system using one-quarter pixel motion compensation, including: applying a filter having negative lobes to a half-way subpixel point between adjacent first and second pixels to generate a one-half filtered pixel value; applying a filter having negative lobes to a one-quarter-way subpixel point between the first and second pixels; and applying a filter having negative lobes to a three-quarter- way subpixel point between the first and second pixels.
59. A method for enhancing image quality in an image encoding system using one-half pixel motion compensation, including applying a filter having negative lobes to a half-way subpixel point between adjacent first and second pixels to generate a one-half filtered pixel value.
60. A method for enhancing image quality in an image encoding system using one-half pixel motion compensation for a luminance channel, including filtering each chrominance channel using one-quarter pixel resolution.
61. The method of claim 60, further including applying a filter having negative lobes to each one-quarter subpixel point between adjacent first and second chrominance pixels.
62. A method for enhancing image quality in an image encoding system using one-quarter pixel motion compensation for a luminance channel, including filtering each chrominance channel using one-eighth pixel resolution.
63. The method of claim 62, further including applying a filter having negative lobes to each one-eighth subpixel point between adjacent first and second chrominance pixels.
64. The method of claims 58, 59, 61 , or 63, wherein the filter having negative lobes is a truncated sine filter.
65. A method for characterizing and correcting the output of an electronic imaging system that produces input images for a video compression system, including: measuring horizontal and vertical color misalignment for pairings of color pixel sensor types ofthe imaging system; measuring noise generated by the color pixel sensor types ofthe imaging system; correcting images generated by the imaging system, before compressing in a video compression system, by: translating color pixels in the images by an amount determined by the measured horizontal and vertical color misalignment; and applying a weighted noise reduction filter to the images, the weighted noise reduction filter having weights that compensate for the amount of any measured noise.
66. A method for characterizing and correcting the ouφut of a film-based imaging system that produces input images for a video compression system, including: determining a film type used for recording a sequence of images; exposing test strips of such film type under a variety of lighting conditions; scanning the exposed test strips through an electronic imaging system having known noise characteristics; measuring noise generated by electronic imaging system during such scanning; and correcting images generated by the film-based imaging system on the same film type, under similar exposure conditions, and scanned by the same electronic imaging system as the test strips, before compressing in a video compression system, by applying a noise reduction filter to the images having weights adjusted for the amount of any measured noise.
67. A method for optimizing conversion of 24 φs film images to video using 3-2 pulldown, including: converting 24 φs film images to digital images using only processing equipment capable of direct 24 φs storage, processing, or communication of such digital images; storing all such digital images in a 24 φs format as a digital image source; performing 3-2 pulldown video conversion on the fly directly from the digital image source using a deterministic frame cadence to create a 3-2 video image sequence; maintaining the deterministic frame cadence for all uses ofthe 3-2 video image sequence; and undoing the deterministic frame cadence and converting the 3-2 video image sequence back to a 24 φs format digital image for storage after use ofthe 3-2 video image sequence.
68. A method for synthesizing a 24 φs moving image from a 72 φs image source, including synthesizing each image frame ofthe 24 φs moving image from three consecutive frames from the 72 φs image source as a weighted average of such frames, wherein the weights for the three frames are in the range of [0.1, 0.8, 0.1] to [0.25, 0.50, 0.25], respectively.
69. The method of claim 68, wherein the weights are approximately [0.1667, 0.6666, 0.1667].
70. A method for synthesizing a 24 φs moving image from a 120 φs image source, including synthesizing each image frame ofthe 24 φs moving image from five consecutive frames from the 120 φs image source as a weighted average of such frames, wherein the weights for the five frames are approximately [0.1, 0.2, 0.4, 0.2, 0.1].
71. A method for synthesizing a 60 φs moving image from a 120 φs image source, including synthesizing each image frame ofthe 60 φs moving image from three consecutive frames from the 120 φs image source as a weighted average of such frames, wherein the weights for the three frames are in the range of [0.1, 0.8, 0.1] to [0.25, 0.50, 0.25], respectively, and overlapping, by one frame, the three consecutive frames used for synthesizing each such image frame with a next three consecutive frames used for synthesizing a next image frame.
72. A method for allocating coding bits in a digital video compression system, including: detecting high compression stress occurring within a selected frame- based unit of video images normally allocated a first constant number of coding bits, such detected unit being a high-stress unit; allocating a second constant number of coding bits greater than the first constant number of coding bits to improve compression ofthe high-stress unit; and compressing at least a remaining part ofthe high-stress unit using the second constant number of coding bits.
73. The method of claim 72, wherein the frame-based unit of video images includes one of a P frame or a Group of Pictures range of frames.
74. The method of claim 72, wherein the second constant number of coding bits is a simple multiple ofthe first constant number of coding bits.
75. The method of claim 72, wherein detecting high compression stress is based upon a rate-control quantization scale factor parameter for the selected frame- based unit of video images.
76. The method of claim 72, including compressing all ofthe high-stress unit using the second constant number of coding bits.
77. A method for improving decoding of compressed digital video information through a decoder having a decoding bit rate and a buffer system, the compressed digital video information being provided from a source at a source bit rate higher than the decoding bit rate, the method including: pre-loading interstitial compressed digital video information from the source into a first portion ofthe buffer system at the source bit rate; concurrently pre-loading program content compressed digital video information from the source into a second portion ofthe buffer system at the source bit rate; selectively changing from the program content compressed digital video information to the interstitial compressed digital video information; and decoding the interstitial compressed digital video information, thereby supporting essentially instantaneous changes in the program content.
78. A method for improving decoding of compressed digital video information through a decoder having a buffer system, an average decoding bit rate, and at least one decoding bit rate higher than the average decoding bit rate, the compressed digital video information being provided from a source at a source bit rate higher than the average decoding bit rate, the method including: pre-loading into a first portion ofthe buffer system at the source bit rate compressed digital video information comprising increased bit rate modules; concurrently pre-loading into a second portion ofthe buffer system at the source bit rate compressed digital video information comprising non-increased bit rate modules; and decoding the contents ofthe second portion ofthe buffer system into a video image at the average decoding bit rate, and decoding the contents ofthe first portion ofthe buffer system into a video image at a decoding bit rate higher than the average decoding bit rate.
79. A method for improving decoding of compressed digital video information through a decoder having a buffer system, an average decoding bit rate, and at least one decoding bit rate higher than the average decoding bit rate, the compressed digital video information being provided from a source at a source bit rate higher than the average decoding bit rate, the method including: pre-loading into a first portion ofthe buffer system at the source bit rate compressed digital video information comprising a compressed enhancement layer; concurrently pre-loading into a second portion ofthe buffer system at the source bit rate compressed digital video information comprising a base layer; and decoding the contents ofthe second portion ofthe buffer system into a video image at the average decoding bit rate, and decoding the contents ofthe first portion ofthe buffer system into a video image at a decoding bit rate higher than the average decoding bit rate.
80. A method for improving the coding efficiency of a video encoding system utilizing discrete cosine transforms (DCT) for coding a base layer and at least one resolution enhancement layer of video images, including: encoding the base layer using DCT blocks each with a first block size; and encoding each resolution enhancement layer using DCT blocks each with a block size that is proportional in size to the first block size as the resolution of such enhancement layer is to the resolution ofthe base layer.
81. The method of claim 80, further including utilizing a subset of a DCT block for an enhancement layer, where such subset corresponds to a DCT block for a lower level enhancement or base layer, to augment the signal-to-noise ratio precision of such DCT block for the lower level enhancement or base layer.
82. A method for determining motion compensation vectors for a base layer and at least one resolution enhancement layer in a video image encoding system, including: encoding the base layer and each resolution enhancement layer using macroblocks sized to cover regions of corresponding pixels within such layers; determining independently, for each macroblock of each base layer and resolution enhancement layer, a number of motion vector subblocks for such macroblock that optimizes a balance between coding prediction quality and a number of bits required to specify an associated set of motion vectors; and determining a set of associated independent motion vectors, one for each ofthe determined number of motion vector subblocks.
83. A method for compressing a video image coding unit, including: applying a plurality of variable length coding tables to each coding unit; selecting the variable length coding table which provides optimum compression for such coding unit; applying the selected variable length coding table to compress such coding unit; and identifying the selected variable length coding table for each such coding unit to a decoder for decompressing such coding unit.
84. The method of claim 83, wherein the coding unit is one of a subframe, a frame, or a group of frames.
85. A method for encoding and decoding video images, including: encoding video images into a first data stream compatible with a basic video compression process and an enhanced video compression process, and a second data stream having constructs compatible only with the enhanced video compression process; decoding only the first data stream on a decoding system compatible with only the basic video compression process; and combining and decoding the first data stream and the second data stream on a decoding system compatible with the enhanced video compression process.
86. The method of claim 85, wherein the basic video compression process and the enhanced video compression process share a common motion-compensated discrete cosine transform structure.
87. The method of claim 85, wherein the basic video compression process is MPEG-2.
88. The method of claim 87, wherein the enhanced video compression process is MPEG-4.
89. A method for motion-compensated encoding of video images in a layered video compression system, including: determining at least one base layer motion vector for a base layer of encoded video images; scaling up each base layer motion vector to the resolution of at least one associated resolution enhancement layer of video information; and determining, for each associated resolution enhancement layer, at least one resolution enhancement layer motion vector, each resolution enhancement layer motion vector corresponding to one ofthe base layer motion vectors, using such one corresponding base layer motion vector as a guide vector to indicate a center point of a restricted search range in such associated resolution enhancement layer for determining such resolution enhancement layer motion vectors.
90. The method of claim 89, further including encoding, for each enhancement layer, only the corresponding resolution enhancement layer motion vectors.
91. The method of claim 89, further including using the vector sum of each resolution enhancement layer motion vector and the corresponding base layer motion vector to provide motion compensation for the enhancement layer associated with such resolution enhancement layer motion vector.
92. A method for compressing a video image, including: downfiltering an initial high resolution image to create a first processed image; generating first motion vectors from the initial high resolution image; compressing the first processed image to create an ouφut base layer; decompressing the output base layer to create a second processed image; expanding the second processed image to create a third processed image; expanding the first processed image to create a fourth processed image; subtracting the third processed image from the initial high resolution image to create a fifth processed image; subtracting the fourth processed image from the initial high resolution image to create a sixth processed image; decreasing the amplitude ofthe sixth processed image to create a seventh processed image; adding the seventh processed image and the fifth processed image to create an eighth processed image; encoding the eighth processed image using the first motion vectors to create an output resolution enhancement layer; decoding the ouφut enhancement layer to create a ninth processed image; adding the ninth processed image and the third processed image to create a tenth processed image; subtracting the initial high resolution image from the tenth processed image to create an eleventh processed image; increasing the amplitude ofthe eleventh processed image to create a twelfth processed image; extracting separate color channels from the twelfth processed image to create a set of thirteenth processed images; encoding the set of thirteenth processed images using the first motion vectors to create a corresponding set of ouφut color resolution enhancement layers; decoding the set of ouφut color enhancement layers to create a set of fourteenth processed images; combining the set of fourteenth processed images to create a fifteenth processed image; decreasing the amplitude ofthe fifteenth processed image to create a sixteenth processed image; adding the sixteenth processed image and the tenth processed image to create a seventeenth processed image; subtract the seventeenth processed image from the initial high resolution image to create an eighteenth processed image; and compressing the eighteenth processed image as ouφut final difference residuals.
93. A method for compressing a video image, including: generating a base layer from an initial high resolution image; generating a first set of motion vectors from a selected image based on the initial high resolution image; generating a first difference image from the initial high resolution image and the base layer; generating a second difference image from the initial high resolution image and a processed copy ofthe initial high resolution image; and generating a resolution enhancement layer from the first and second difference images and the first set of motion vectors.
94. The method of claim 93, further including generating at least one color resolution enhancement layer for at least one selected color.
95. The method of claim 93, further including generating final difference residuals.
96. The method of claim 95, further including encoding the final difference residuals.
PCT/US2001/011204 2000-04-07 2001-04-06 Enhanced temporal and resolution layering in advanced television WO2001077871A1 (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
JP2001574651A JP4352105B2 (en) 2000-04-07 2001-04-06 Advanced television with enhanced temporal and resolution stratification
AU2001251386A AU2001251386A1 (en) 2000-04-07 2001-04-06 Enhanced temporal and resolution layering in advanced television
EP01924762A EP1279111A4 (en) 2000-04-07 2001-04-06 Enhanced temporal and resolution layering in advanced television
CA002406459A CA2406459C (en) 2000-04-07 2001-04-06 Enhanced temporal and resolution layering in advanced television

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US09/545,233 US6728317B1 (en) 1996-01-30 2000-04-07 Moving image compression quality enhancement using displacement filters with negative lobes
US09/545,233 2000-04-07

Publications (1)

Publication Number Publication Date
WO2001077871A1 true WO2001077871A1 (en) 2001-10-18

Family

ID=24175400

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2001/011204 WO2001077871A1 (en) 2000-04-07 2001-04-06 Enhanced temporal and resolution layering in advanced television

Country Status (5)

Country Link
EP (1) EP1279111A4 (en)
JP (1) JP4352105B2 (en)
AU (1) AU2001251386A1 (en)
CA (1) CA2406459C (en)
WO (1) WO2001077871A1 (en)

Cited By (47)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1460846A1 (en) * 2001-12-13 2004-09-22 Sony Corporation Image signal processing apparatus and processing method
WO2004095829A1 (en) * 2003-04-10 2004-11-04 Thomson Licensing S.A. Technique for simulating film grain on encoded video
WO2004102963A1 (en) 2003-05-16 2004-11-25 Sony Corporation Motion correction device and method
EP1501312A2 (en) * 2003-07-18 2005-01-26 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP1511320A1 (en) * 2003-09-01 2005-03-02 Matsushita Electric Industrial Co., Ltd. Film grain encoding
EP1526730A1 (en) * 2002-04-17 2005-04-27 Matsushita Electric Industrial Co., Ltd. Image conversion device and image conversion method
EP1536643A1 (en) * 2002-04-17 2005-06-01 Matsushita Electric Industrial Co., Ltd. Image conversion device and image conversion method
WO2005076599A1 (en) * 2004-02-03 2005-08-18 Koninklijke Philips Electronics N.V. Changing the aspect ratio of images to be displayed on a screen
FR2872664A1 (en) * 2004-07-01 2006-01-06 Nextream France Sa DEVICE AND METHOD FOR PRE-TRAITEMEBNT BEFORE ENCODING A SEQUENCE OF VIDEO IMAGES
WO2006008609A1 (en) * 2004-07-12 2006-01-26 Nokia Corporation System and method for motion prediction in scalable video coding
WO2006020532A2 (en) * 2004-08-09 2006-02-23 Pinnacle Systems, Inc. Fast area-selected filtering for pixel-noise and analog artifacts reduction
JP2006509421A (en) * 2002-12-03 2006-03-16 トムソン ライセンシング Digital video disc for standard and high definition video formats on a single disc
FR2876860A1 (en) * 2004-10-20 2006-04-21 Thomson Licensing Sa METHOD FOR HIERARCHIC ENCODING OF VIDEO IMAGES
FR2876861A1 (en) * 2004-10-20 2006-04-21 Thomson Licensing Sa Video image e.g. digital television video image, encoding method for providing encoded data stream, involves encoding standard definition format image for parts common to zoomed image to provide encoded data to upper encoded stream layer
WO2006058921A1 (en) * 2004-12-03 2006-06-08 Thomson Licensing Method for scalable video coding
FR2879066A1 (en) * 2004-12-03 2006-06-09 Thomson Licensing Sa Coding data inheriting method from images with lower resolution, by assigning mode and motion data of zoomed BR block to HR block if number of zoomed BR blocks is 1
WO2006064422A1 (en) * 2004-12-13 2006-06-22 Koninklijke Philips Electronics N.V. Scalable picture encoding
EP1737240A2 (en) * 2005-06-21 2006-12-27 Thomson Licensing Method for scalable image coding or decoding
JP2007514336A (en) * 2003-05-15 2007-05-31 トムソン ライセンシング Method and apparatus for representing image graininess by one or more parameters
WO2007063912A1 (en) 2005-11-29 2007-06-07 Matsushita Electric Industrial Co., Ltd. Reproduction device
WO2007080477A3 (en) * 2006-01-10 2007-10-25 Nokia Corp Switched filter up-sampling mechanism for scalable video coding
US7366242B2 (en) 1996-01-30 2008-04-29 Dolby Laboratories Licensing Corporation Median filter combinations for video noise reduction
JP2008530839A (en) * 2005-02-07 2008-08-07 トムソン ライセンシング Method and apparatus for reproducing a video signal and one or more audio signals related to audio / video data based on a 24 Hz frame frequency video signal
EP2129108A1 (en) * 2006-12-18 2009-12-02 Sony Corporation Imaging device and method, recording device and method, and reproduction device and method
US7680356B2 (en) 2003-10-14 2010-03-16 Thomson Licensing Technique for bit-accurate comfort noise addition
US7738721B2 (en) 2003-08-29 2010-06-15 Thomson Licensing Method and apparatus for modeling film grain patterns in the frequency domain
US7738722B2 (en) 2004-10-21 2010-06-15 Thomson Licensing Technique for adaptive de-blocking of block-based film grain patterns
EP2254339A2 (en) 2002-06-28 2010-11-24 Dolby Laboratories Licensing Corporation Improved interpolation of compressed video frames
US7843508B2 (en) 2002-07-23 2010-11-30 Mediostream, Inc. Method and system for direct recording of video information onto a disk medium
US7852409B2 (en) 2004-11-16 2010-12-14 Thomson Licensing Bit-accurate seed initialization for pseudo-random number generators used in a video system
US7889939B2 (en) 2003-09-23 2011-02-15 Thomson Licensing Technique for simulating film grain using frequency filtering
US7945106B2 (en) 2003-09-23 2011-05-17 Thomson Licensing Method for simulating film grain by mosaicing pre-computer samples
CN102075755A (en) * 2005-03-18 2011-05-25 夏普株式会社 Methods and systems for picture up-sampling
US8014558B2 (en) 2004-10-18 2011-09-06 Thomson Licensing Methods, apparatus and system for film grain simulation
US8150206B2 (en) 2004-03-30 2012-04-03 Thomson Licensing Method and apparatus for representing image granularity by one or more parameters
US8238613B2 (en) 2003-10-14 2012-08-07 Thomson Licensing Technique for bit-accurate film grain simulation
US8345762B2 (en) 2005-02-18 2013-01-01 Thomson Licensing Method for deriving coding information for high resolution pictures from low resolution pictures and coding and decoding devices implementing said method
US8446956B2 (en) 2006-01-05 2013-05-21 Thomson Licensing Inter-layer motion prediction method using resampling
WO2014137920A1 (en) * 2013-03-05 2014-09-12 Qualcomm Incorporated Inter-layer reference picture construction for spatial scalability with different aspect ratios
US9098916B2 (en) 2004-11-17 2015-08-04 Thomson Licensing Bit-accurate film grain simulation method based on pre-computed transformed coefficients
US9117261B2 (en) 2004-11-16 2015-08-25 Thomson Licensing Film grain SEI message insertion for bit-accurate simulation in a video system
US9167266B2 (en) 2006-07-12 2015-10-20 Thomson Licensing Method for deriving motion for high resolution pictures from motion data of low resolution pictures and coding and decoding devices implementing said method
US9177364B2 (en) 2004-11-16 2015-11-03 Thomson Licensing Film grain simulation method based on pre-computed transform coefficients
EP3340625A4 (en) * 2015-08-19 2019-01-23 Sony Corporation Transmission device, transmission method, reception device, and reception method
US10715834B2 (en) 2007-05-10 2020-07-14 Interdigital Vc Holdings, Inc. Film grain simulation based on pre-computed transform coefficients
CN113316001A (en) * 2021-05-25 2021-08-27 上海哔哩哔哩科技有限公司 Video alignment method and device
WO2022251383A1 (en) * 2021-05-26 2022-12-01 Qualcomm Incorporated High quality ui element boundaries using masks in temporally interpolated frames

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008515328A (en) * 2004-10-18 2008-05-08 サムスン エレクトロニクス カンパニー リミテッド Video coding and decoding method using inter-layer filtering, video encoder and decoder
EP2002658A2 (en) * 2006-03-23 2008-12-17 Koninklijke Philips Electronics N.V. Coding device and method for encoding movie data

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5742343A (en) * 1993-07-13 1998-04-21 Lucent Technologies Inc. Scalable encoding and decoding of high-resolution progressive video
US5828788A (en) * 1995-06-29 1998-10-27 Thomson Multimedia, S.A. System for processing data in variable segments and with variable data resolution
US5852565A (en) * 1996-01-30 1998-12-22 Demografx Temporal and resolution layering in advanced television
US6028634A (en) * 1995-10-27 2000-02-22 Kabushiki Kaisha Toshiba Video encoding and decoding apparatus

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5742343A (en) * 1993-07-13 1998-04-21 Lucent Technologies Inc. Scalable encoding and decoding of high-resolution progressive video
US5828788A (en) * 1995-06-29 1998-10-27 Thomson Multimedia, S.A. System for processing data in variable segments and with variable data resolution
US6028634A (en) * 1995-10-27 2000-02-22 Kabushiki Kaisha Toshiba Video encoding and decoding apparatus
US5852565A (en) * 1996-01-30 1998-12-22 Demografx Temporal and resolution layering in advanced television
US5988863A (en) * 1996-01-30 1999-11-23 Demografx Temporal and resolution layering in advanced television

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP1279111A4 *

Cited By (109)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
USRE44235E1 (en) 1996-01-30 2013-05-21 Dolby Laboratories Licensing Corporation Enhancing image quality in an image system
USRE43360E1 (en) 1996-01-30 2012-05-08 Dolby Laboratories Licensing Corporation Enhancing image quality in an image system
USRE45082E1 (en) 1996-01-30 2014-08-19 Dolby Laboratories Licensing Corporation Enhancing image quality in an image system
US7366242B2 (en) 1996-01-30 2008-04-29 Dolby Laboratories Licensing Corporation Median filter combinations for video noise reduction
US8995528B2 (en) 2001-07-11 2015-03-31 Dolby Laboratories Licensing Corporation Switch-select single frame reference
US8942285B2 (en) 2001-07-11 2015-01-27 Dolby Laboratories Licensing Corporation Motion compensation filtering in an image system
US8767829B2 (en) 2001-07-11 2014-07-01 Dolby Laboratories Licensing Corporation Switch-select single frame reference
EP1460846A1 (en) * 2001-12-13 2004-09-22 Sony Corporation Image signal processing apparatus and processing method
EP1460846A4 (en) * 2001-12-13 2009-12-09 Sony Corp Image signal processing apparatus and processing method
EP1526730A1 (en) * 2002-04-17 2005-04-27 Matsushita Electric Industrial Co., Ltd. Image conversion device and image conversion method
EP1536643A4 (en) * 2002-04-17 2005-09-07 Matsushita Electric Ind Co Ltd Image conversion device and image conversion method
EP1526730A4 (en) * 2002-04-17 2005-09-07 Matsushita Electric Ind Co Ltd Image conversion device and image conversion method
US7446815B2 (en) 2002-04-17 2008-11-04 Matsushita Electric Industrial Co., Ltd. Image conversion device and image conversion method
EP1536643A1 (en) * 2002-04-17 2005-06-01 Matsushita Electric Industrial Co., Ltd. Image conversion device and image conversion method
US7388617B2 (en) 2002-04-17 2008-06-17 Matsushita Electric Industrial Co., Ltd. Image conversion device and image conversion method
EP2458864A2 (en) 2002-06-28 2012-05-30 Dolby Laboratories Licensing Corporation Improved interpolation of compressed video frames
EP2254339A2 (en) 2002-06-28 2010-11-24 Dolby Laboratories Licensing Corporation Improved interpolation of compressed video frames
EP2458863A2 (en) 2002-06-28 2012-05-30 Dolby Laboratories Licensing Corporation Improved interpolation of compressed video frames
US7843508B2 (en) 2002-07-23 2010-11-30 Mediostream, Inc. Method and system for direct recording of video information onto a disk medium
US8619188B2 (en) 2002-07-23 2013-12-31 Mediostream, Inc. Method and system for direct recording of video information onto a disk medium
JP2006509421A (en) * 2002-12-03 2006-03-16 トムソン ライセンシング Digital video disc for standard and high definition video formats on a single disc
JP4751614B2 (en) * 2002-12-03 2011-08-17 トムソン ライセンシング Digital video disc for standard and high definition video formats on a single disc
US7912125B2 (en) 2002-12-03 2011-03-22 Thomson Licensing Hybrid scalable encoder, method and media for standard definition and high-definition video formats on a single-disc
US7899113B2 (en) 2003-04-10 2011-03-01 Thomson Licensing Technique for simulating film grain on encoded video
KR100987666B1 (en) * 2003-04-10 2010-10-13 톰슨 라이센싱 Technique for simulating film grain on encoded video
WO2004095829A1 (en) * 2003-04-10 2004-11-04 Thomson Licensing S.A. Technique for simulating film grain on encoded video
JP2011071998A (en) * 2003-05-15 2011-04-07 Thomson Licensing Method and apparatus for representing image granularity by one or more parameters
JP2007515846A (en) * 2003-05-15 2007-06-14 トムソン ライセンシング Method and apparatus for representing image graininess by one or more parameters
US7742655B2 (en) 2003-05-15 2010-06-22 Thomson Licensing Method and apparatus for representing image granularity by one or more parameters
JP2007514336A (en) * 2003-05-15 2007-05-31 トムソン ライセンシング Method and apparatus for representing image graininess by one or more parameters
EP1627360B1 (en) * 2003-05-15 2018-10-03 Dolby International AB Method and apparatus for representing image granularity by one or more parameters
WO2004102963A1 (en) 2003-05-16 2004-11-25 Sony Corporation Motion correction device and method
EP1513344A4 (en) * 2003-05-16 2009-10-28 Sony Corp Motion correction device and method
EP1513344A1 (en) * 2003-05-16 2005-03-09 Sony Corporation Motion correction device and method
EP1501312A2 (en) * 2003-07-18 2005-01-26 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP2323400A3 (en) * 2003-07-18 2011-08-10 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US9042443B2 (en) 2003-07-18 2015-05-26 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US9042442B2 (en) 2003-07-18 2015-05-26 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US9706215B2 (en) 2003-07-18 2017-07-11 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP2323396A3 (en) * 2003-07-18 2011-08-10 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US9961356B2 (en) 2003-07-18 2018-05-01 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP2323397A3 (en) * 2003-07-18 2011-08-10 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP1959689A3 (en) * 2003-07-18 2010-05-12 Samsung Electronics Co., Ltd Image encoding and decoding apparatus and method
US9706216B2 (en) 2003-07-18 2017-07-11 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP2323395A3 (en) * 2003-07-18 2011-08-10 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US10602172B2 (en) 2003-07-18 2020-03-24 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US9729892B2 (en) 2003-07-18 2017-08-08 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP2323401A3 (en) * 2003-07-18 2011-08-10 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
EP1501312A3 (en) * 2003-07-18 2005-10-26 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US8345748B2 (en) 2003-07-18 2013-01-01 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US8270474B2 (en) 2003-07-18 2012-09-18 Samsung Electronics Co., Ltd. Image encoding and decoding apparatus and method
US7738721B2 (en) 2003-08-29 2010-06-15 Thomson Licensing Method and apparatus for modeling film grain patterns in the frequency domain
EP1511320A1 (en) * 2003-09-01 2005-03-02 Matsushita Electric Industrial Co., Ltd. Film grain encoding
US7889939B2 (en) 2003-09-23 2011-02-15 Thomson Licensing Technique for simulating film grain using frequency filtering
US7945106B2 (en) 2003-09-23 2011-05-17 Thomson Licensing Method for simulating film grain by mosaicing pre-computer samples
US8238613B2 (en) 2003-10-14 2012-08-07 Thomson Licensing Technique for bit-accurate film grain simulation
US7680356B2 (en) 2003-10-14 2010-03-16 Thomson Licensing Technique for bit-accurate comfort noise addition
WO2005076599A1 (en) * 2004-02-03 2005-08-18 Koninklijke Philips Electronics N.V. Changing the aspect ratio of images to be displayed on a screen
US8150206B2 (en) 2004-03-30 2012-04-03 Thomson Licensing Method and apparatus for representing image granularity by one or more parameters
WO2006003102A1 (en) * 2004-07-01 2006-01-12 Thomson Licensing S.A. Pre-processing device and method before encoding of a video image sequence
KR101160718B1 (en) * 2004-07-01 2012-06-28 톰슨 라이센싱 Pre-processing device and method before encoding of a video image sequence
FR2872664A1 (en) * 2004-07-01 2006-01-06 Nextream France Sa DEVICE AND METHOD FOR PRE-TRAITEMEBNT BEFORE ENCODING A SEQUENCE OF VIDEO IMAGES
US8537904B2 (en) 2004-07-01 2013-09-17 Thomson Licensing Pre-processing device and method before encoding of a video image sequence
WO2006008609A1 (en) * 2004-07-12 2006-01-26 Nokia Corporation System and method for motion prediction in scalable video coding
WO2006020532A3 (en) * 2004-08-09 2006-04-06 Pinnacle Systems Inc Fast area-selected filtering for pixel-noise and analog artifacts reduction
WO2006020532A2 (en) * 2004-08-09 2006-02-23 Pinnacle Systems, Inc. Fast area-selected filtering for pixel-noise and analog artifacts reduction
US8014558B2 (en) 2004-10-18 2011-09-06 Thomson Licensing Methods, apparatus and system for film grain simulation
US9117260B2 (en) 2004-10-18 2015-08-25 Thomson Licensing Methods for determining block averages for film grain simulation
US9953401B2 (en) 2004-10-18 2018-04-24 Thomson Licensing Apparatus and system for determining block averages for film grain simulation
FR2876860A1 (en) * 2004-10-20 2006-04-21 Thomson Licensing Sa METHOD FOR HIERARCHIC ENCODING OF VIDEO IMAGES
US8306119B2 (en) 2004-10-20 2012-11-06 Thomson Licensing Method for hierarchically coding video images
FR2876861A1 (en) * 2004-10-20 2006-04-21 Thomson Licensing Sa Video image e.g. digital television video image, encoding method for providing encoded data stream, involves encoding standard definition format image for parts common to zoomed image to provide encoded data to upper encoded stream layer
WO2006059021A1 (en) * 2004-10-20 2006-06-08 Thomson Licensing Method for encoding video images having different nonproportional formats
WO2006042990A1 (en) * 2004-10-20 2006-04-27 Thomson Licensing Method for hierarchically coding video images
US7738722B2 (en) 2004-10-21 2010-06-15 Thomson Licensing Technique for adaptive de-blocking of block-based film grain patterns
US7852409B2 (en) 2004-11-16 2010-12-14 Thomson Licensing Bit-accurate seed initialization for pseudo-random number generators used in a video system
US9117261B2 (en) 2004-11-16 2015-08-25 Thomson Licensing Film grain SEI message insertion for bit-accurate simulation in a video system
US9177364B2 (en) 2004-11-16 2015-11-03 Thomson Licensing Film grain simulation method based on pre-computed transform coefficients
US9098916B2 (en) 2004-11-17 2015-08-04 Thomson Licensing Bit-accurate film grain simulation method based on pre-computed transformed coefficients
WO2006058921A1 (en) * 2004-12-03 2006-06-08 Thomson Licensing Method for scalable video coding
FR2879066A1 (en) * 2004-12-03 2006-06-09 Thomson Licensing Sa Coding data inheriting method from images with lower resolution, by assigning mode and motion data of zoomed BR block to HR block if number of zoomed BR blocks is 1
WO2006064422A1 (en) * 2004-12-13 2006-06-22 Koninklijke Philips Electronics N.V. Scalable picture encoding
US8244094B2 (en) 2005-02-07 2012-08-14 Thomson Licensing Method and apparatus for replaying a video signal and one or more audio signals related to audio/video data that are based on a 24Hz frame frequency video signal
JP2008530839A (en) * 2005-02-07 2008-08-07 トムソン ライセンシング Method and apparatus for reproducing a video signal and one or more audio signals related to audio / video data based on a 24 Hz frame frequency video signal
US8345762B2 (en) 2005-02-18 2013-01-01 Thomson Licensing Method for deriving coding information for high resolution pictures from low resolution pictures and coding and decoding devices implementing said method
CN102075755A (en) * 2005-03-18 2011-05-25 夏普株式会社 Methods and systems for picture up-sampling
EP1737240A2 (en) * 2005-06-21 2006-12-27 Thomson Licensing Method for scalable image coding or decoding
WO2006136568A2 (en) * 2005-06-21 2006-12-28 Thomson Licensing Method for scalable image coding or decoding
WO2006136568A3 (en) * 2005-06-21 2007-04-26 Thomson Licensing Method for scalable image coding or decoding
EP1737240A3 (en) * 2005-06-21 2007-03-14 Thomson Licensing Method for scalable image coding or decoding
EP1956830A1 (en) * 2005-11-29 2008-08-13 Matsushita Electric Industrial Co., Ltd. Reproduction device
EP1956830A4 (en) * 2005-11-29 2010-09-29 Panasonic Corp Reproduction device
WO2007063912A1 (en) 2005-11-29 2007-06-07 Matsushita Electric Industrial Co., Ltd. Reproduction device
US8351756B2 (en) 2005-11-29 2013-01-08 Panasonic Corporation Reproduction device
US8446956B2 (en) 2006-01-05 2013-05-21 Thomson Licensing Inter-layer motion prediction method using resampling
WO2007080477A3 (en) * 2006-01-10 2007-10-25 Nokia Corp Switched filter up-sampling mechanism for scalable video coding
US9167266B2 (en) 2006-07-12 2015-10-20 Thomson Licensing Method for deriving motion for high resolution pictures from motion data of low resolution pictures and coding and decoding devices implementing said method
US8102436B2 (en) 2006-12-18 2012-01-24 Sony Corporation Image-capturing apparatus and method, recording apparatus and method, and reproducing apparatus and method
EP2129108A4 (en) * 2006-12-18 2011-10-26 Sony Corp Imaging device and method, recording device and method, and reproduction device and method
EP2129108A1 (en) * 2006-12-18 2009-12-02 Sony Corporation Imaging device and method, recording device and method, and reproduction device and method
US10715834B2 (en) 2007-05-10 2020-07-14 Interdigital Vc Holdings, Inc. Film grain simulation based on pre-computed transform coefficients
JP2016513913A (en) * 2013-03-05 2016-05-16 クゥアルコム・インコーポレイテッドＱｕａｌｃｏｍｍ Ｉｎｃｏｒｐｏｒａｔｅｄ Construction of reference pictures between layers for spatial scalability with different aspect ratios
US10284842B2 (en) 2013-03-05 2019-05-07 Qualcomm Incorporated Inter-layer reference picture construction for spatial scalability with different aspect ratios
WO2014137920A1 (en) * 2013-03-05 2014-09-12 Qualcomm Incorporated Inter-layer reference picture construction for spatial scalability with different aspect ratios
EP3340625A4 (en) * 2015-08-19 2019-01-23 Sony Corporation Transmission device, transmission method, reception device, and reception method
CN113316001A (en) * 2021-05-25 2021-08-27 上海哔哩哔哩科技有限公司 Video alignment method and device
CN113316001B (en) * 2021-05-25 2023-04-11 上海哔哩哔哩科技有限公司 Video alignment method and device
WO2022251383A1 (en) * 2021-05-26 2022-12-01 Qualcomm Incorporated High quality ui element boundaries using masks in temporally interpolated frames
US11587208B2 (en) 2021-05-26 2023-02-21 Qualcomm Incorporated High quality UI elements with frame extrapolation

Also Published As

Publication number Publication date
JP4352105B2 (en) 2009-10-28
EP1279111A1 (en) 2003-01-29
CA2406459C (en) 2006-06-06
JP2003531514A (en) 2003-10-21
EP1279111A4 (en) 2005-03-23
CA2406459A1 (en) 2001-10-18
AU2001251386A1 (en) 2001-10-23

Similar Documents

Publication Publication Date Title
US6728317B1 (en) Moving image compression quality enhancement using displacement filters with negative lobes
CA2406459C (en) Enhanced temporal and resolution layering in advanced television
KR100481572B1 (en) Time and Resolution Layering in ATVs
JP4266389B2 (en) General purpose video disc recording and playback using motion signals for high quality playback of non-film sources
US7280155B2 (en) Method and system for converting interlaced formatted video to progressive scan video
US6862372B2 (en) System for and method of sharpness enhancement using coding information and local spatial features
US6023295A (en) ADPCM recompression and decompression of a data stream of a video image and differential variance estimator
WO1993010628A1 (en) Motion filter for digital television system
US6873657B2 (en) Method of and system for improving temporal consistency in sharpness enhancement for a video signal
EP1506525B1 (en) System for and method of sharpness enhancement for coded digital video
US20070086666A1 (en) Compatible interlaced sdtv and progressive hdtv
Demos Temporal and resolution layering in advanced television
US6816553B1 (en) Coding method for an image signal
JPH10112857A (en) Method and device for coding and decoding image signal
Reimers et al. JPEG and MPEG Source Coding of Video Signals
van Rooy The LDK2000 Multistandard Camera, and the Evolution to DTV
JP2004515133A (en) Decompression of encoded video
Bayrakeri Scalable video coding using spatiotemporal interpolation
Drury Video preprocessing in MPEG-2 compression systems

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT TZ UA UG US UZ VN YU ZA ZW

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
WWE Wipo information: entry into national phase

Ref document number: 2406459

Country of ref document: CA

ENP Entry into the national phase

Ref country code: JP

Ref document number: 2001 574651

Kind code of ref document: A

Format of ref document f/p: F

WWE Wipo information: entry into national phase

Ref document number: 2001924762

Country of ref document: EP

WWP Wipo information: published in national office

Ref document number: 2001924762

Country of ref document: EP

WWW Wipo information: withdrawn in national office

Ref document number: 2001924762

Country of ref document: EP