WO2001091473A1 - Method for testing video sequences - Google Patents

Method for testing video sequences Download PDF

Info

Publication number
WO2001091473A1
WO2001091473A1 PCT/GB2001/002374 GB0102374W WO0191473A1 WO 2001091473 A1 WO2001091473 A1 WO 2001091473A1 GB 0102374 W GB0102374 W GB 0102374W WO 0191473 A1 WO0191473 A1 WO 0191473A1
Authority
WO
WIPO (PCT)
Prior art keywords
frame
video
sequence
pixels
quality
Prior art date
Application number
PCT/GB2001/002374
Other languages
French (fr)
Inventor
Alexandre Jean Bourret
Original Assignee
British Telecommunications Public Limited Company
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
Application filed by British Telecommunications Public Limited Company filed Critical British Telecommunications Public Limited Company
Priority to US10/275,474 priority Critical patent/US7233348B2/en
Priority to CA2408435A priority patent/CA2408435C/en
Priority to EP01931970A priority patent/EP1290900A1/en
Priority to AU58652/01A priority patent/AU5865201A/en
Publication of WO2001091473A1 publication Critical patent/WO2001091473A1/en

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N17/00Diagnosis, testing or measuring for television systems or their details
    • H04N17/004Diagnosis, testing or measuring for television systems or their details for digital television systems

Definitions

  • This invention relates to a test method that is can assess the quality of video sequences, in particular digitally encoded video signals.
  • compression algorithms such as H263 or MPEG rely on a set of well known methods. The first of these is to suppress the information present in each frame that is not perceptually relevant. The result of this operation may not be perceptible to the end user, perceptible but not perceptually relevant, or disruptive to the information content of the frame. In most cases, the effect of the information suppression will be to modify the appearance of textured objects, giving a greater repetitiveness or simplicity.
  • a method of comparing two frames from a video sequence comprising the steps of;
  • step (iv) repeating step (iii) for all pixels in a test area of the first frame.
  • results of the comparison stage indicate a large number of substantially equal pixels then a dropped frame may be inferred. If the results of the comparison stage indicate a small number of substantially equal pixels then a scene cut may be inferred.
  • Figure 1 shows a reference video frame
  • Figure 2 shows the video frame of Figure 1 after having been compressed
  • Figure 3 shows a schematic depiction of apparatus for performing a method according to the present invention
  • Figure 4 shows the variation in unchanged pixel proportion with time for a number of different compression schemes
  • Figure 5 shows the correlation of predicted and subjective quality scores for a number of video sequences
  • Figure 6 shows the correlation of predicted and subjective quality scores for a number of subsets of video sequences
  • this simplification operation is done within blocks of pixels, most often of 8x8 pixels.
  • disparities tend to appear between adjacent regions, making the block structure of the image analysis visible.
  • This effect referred as blockiness, can easily be measured (see “Image quality measure based on local visual properties", P Franti, Picture Coding Symposium 97, 10-1 2 Sept. 1997, no.143, pp217-20, or M Ghanbari, "Blockiness Detection for MPEG2-coded video", IEEE Signal Processing Letters, 7:8 (August 2000), pp 21 3-21 5) even without having access to the reference sequence.
  • a second such method used to reduce bandwidth requirements for video transmissions is movement estimation.
  • This technique involves analysing the motion present between one frame and the next frame in the sequence, so that movement vectors can be transmitted to the decoder in order to reconstruct the sequence from a starting frame.
  • motion estimation is not used between every frame. Instead, a temporal pattern can be used, as in the MPEG-2 compression scheme, where original, interpolated and predicted frames are interleaved in a predefined sequence.
  • the third method used is to reduce the frame rate of the sequence. This is used for lower bitrate applications such as videoconferencing or video over IP, and is simply done by dropping one or more frames from a sequence.
  • motion present in a sequence is an important indicator of the image quality. Furthermore, the motion present in a sequence can be determined without the help of the original sequence, enabling the creation of non intrusive assessment systems that can estimate the quality of video systems.
  • FIG. 3 shows video capture unit 10, which is suitable for capturing a series of frames from a video sequence. Each frame from the series is sent to blockiness. detector 40 and also each frame is transmitted to the comparator 20, along with the subsequent frame from the video sequence. Both the blockiness detector 40 and the comparator 20 are connected to a data processing unit 30, so that outputs from the blockiness detector and the comparator can be passed to the data processing unit.
  • the blockiness detector 40 analyses the pixels of each video frame received from the video capture unit.
  • the video compression technique divides the video image into a number of blocks and a side effect of simplifying the content of each block is to increase the disparity between the blocks, making them clearly visible in the reconstructed image.
  • Different approaches have been tried to detect blockiness (see Franti or Ghanbari op cit) but the preferred technique for detecting blockiness for use in the present invention is to compare the variation of a pixel property (for example the luminance) every 8 pixels with the sum of variations within these 8 pixels.
  • a ratio indicating a higher activity every 8 pixels than within 8 pixel blocks triggers an alarm as it tends to indicate an abrupt change in the pixel property in the transition from a block of very similar values to the next block. Alarms are counted across the whole frame, and the sum is used as an indicator of blockiness for the image. These blockiness indicator values are then transmitted to the data processing unit, along with a reference to the frame to which they correspond.
  • the two video frames are then sent to comparator 20 which compares the two frames pixel by pixel using one of the properties of a pixel as a basis for the comparison.
  • the results of the comparator are fed to data processing unit 30.
  • the comparator counts all those pixels where there has been no significant change in the pixel property (i.e. the luminance of the pixel in the subsequent frame is within a pre-determined threshold from the luminance of the pixel in the first frame) and determines the proportion of pixels for which the pixel property has not changed.
  • the proportion of pixels in which there has been no change of pixel property and/or the variation in the proportion of pixels in which there has been no change of pixel property can be used to assess the quality of the video compression.
  • Data processing unit 30 stores the pixel proportion values along with a reference to the frame to which they correspond.
  • Most video capture systems introduce some noise into the image. For instance, the luminance of a given pixel produced by a video camera which is capturing a stationary image, for example background scenery, will vary in time.
  • the video compression rate is increased, i.e. the bandwidth of the transmitted video stream is decreased, the variations of a given pixel tend to be suppressed, and the pixel properties (for example chrominance, luminance, etc.) tend towards a constant value.
  • the pixel properties for example chrominance, luminance, etc.
  • Figure 4 shows the variation of the proportion of unchanged pixels for the reference signal
  • Figure 4b shows the variation of the proportion of unchanged pixels when that signal has been compressed at 4Mb/s
  • Figure 4c shows the variation of the proportion of unchanged pixels when the reference signal has been compressed at 1 Mb/s.
  • the data processing unit has the facility to calculate a value which gives an indication of the variation of pixel proportion values# e.g. by performing a fast Fourier transform (FFT) of the pixel proportion values, calculating the derivative of the pixel proportion value, etc.
  • FFT fast Fourier transform
  • the preferred indicator is to measure the average and standard deviation of the proportion over a 500m window.
  • this may be an indicator that the video sequence is being transmitted across an unreliable transport mechanism (in this case if a video sequence is being transmitted 'live' across a communications network, it may be possible to send a signal to the transmitting codec to decrease the frame rate in use (if the frame rate received is less than the frame rate in use) or to switch to a transport mechanism having a greater quality of service.
  • These three indicators, the amount of blockiness, the proportion of unchanging pixels and the ratio of dropped frames can be used to generate a model that can be used to assess the quality of a received video sequence in a non-intrusive manner, that is without needing to insert any test signals, or without having access to the initial video sequence.
  • a matrix was formed from each of the three indicators and a time-dependent variation of each indicator.
  • the non- intrusive model was trained and calibrated using linear-regression techniques to assess the qualities of a number of known video sequences which had previously been subjectively judged (that is, the quality of the sequences has been assessed by a number of human viewers).
  • the non-intrusive model by design, is looking for known artefacts introduced by the codec in use. For this reason, the characteristics of the codec in use must be known and the model trained for the particular configuration being assessed. This requirement is incompatible with applications such as assessment of quality improvement of compression algorithms or tests on new coding schemes for instance.
  • the model could easily be tricked by hiding such artefacts, without improving the overall quality (the most obvious example would be the use of a blockiness smother algorithm, as described by Ghanbari op cit).
  • this approach can be particularly useful to telecom operators who are not changing coding algorithms on daily basis, but who want to monitor the performance of their network.
  • the first database used to calibrate and train the non-intrusive model is the series of sequences used for the VQEG contests. It consists of a number of sequences which are repeated but at different levels of compression. The resolution used in the sequences and the levels of degradation are consistent with broadcast applications (for the purpose of this calibration, some degradations are discarded, because they are based on analogue techniques.
  • the second database relates to videoconference applications.
  • the format used is either CIF or QCIF, the codec used to generate the degradations are H263 or MPEG4, and IP or UMTS error pattern have been inserted into the data stream before the decoding stage. The main difference between the two databases lies in the way the sequences have been previously assessed.
  • the opinion scores of the VQEG sequences which are about 10 seconds long in average, have been taken at the end of each presentation.
  • the second database which contains longer sequences and shows rapid variations in quality, has been assessed continuously. As test material is complicated to produce, the databases were split into a calibration set and a test set.
  • test set of each database were processed through the model, which produces a set of parameters for each frame it contains (blockiness, pixel variation and dropped frame ratio).
  • An extra set of parameters is derived from these initial data, by computing their mean and mean and standard variation on blocks of one second. These account for the temporal variations, which are linked to the compression rate.
  • a linear regression is used to obtain the best combination between the extracted parameters and the mean opinion score as previously measured for the given sequences. The combination is then tried with on the parameters extracted for the rest of the database. The produced is compared with the measured mean opinion score, giving us an indicator of the model performance.
  • the model was also calibrated using a subset of the Eurescom Aquavit database.
  • This bank of subjectively assessed sequences represents a variety of codec and compression rate at levels compatible with multimedia applications and
  • Figure 5 shows the correlation between the predicted score generated by the non- intrusive model and the mean subjective score for the test set of sequences.
  • Figure 6a shows the variation of correlation between predicted score and subjective score with the frame rate used in the sequence and
  • Figure 6b shows the variation of correlation between predicted score and subjective score with the bandwidth used to code the sequence.
  • Profiles can be incorporated into the model, according to the type of video sequences processed. Requirements and expectations can be very different according to the type of service or images transmitted. At the equivalent level of quality, a video sequence can be acceptable in video conference but not in broadcast. The bandwidth requirement are much higher for sport than head and shoulder sequences. Profiles take these aspects into account before to make a judgement about the quality.
  • the measurement tool can be used in different ways. By providing a means to measure the quality of refreshment in video sequences, it allows comparison between a reference signal and the same signal once gone through by the coding and decoding process. But it can also provides useful information about a sequence without the help of any reference.
  • processing power is limited (for example if a real-time assessment is required or the model is being incorporated into a device with limited capabilities then a less-sophisticated model can be derived by using fewer parameters; the nature of the system to be measured will influence which quality indicator, or its time dependence, can be excluded from the model and the optimal combination can easily be determined by experimentation and further application of linear regression techniques. It will be readily understood that the present invention could be implemented in software alone, i.e.
  • a general purpose computer such as personal computer (which may be supplied on a data carrier such as, for example a floppy disc of a CD-ROM), in hardware alone or in a combination of hardware and software (for example programmable DSPs) and that the current invention should not be limited by ' the form of -the implementation.
  • a general purpose computer such as personal computer (which may be supplied on a data carrier such as, for example a floppy disc of a CD-ROM), in hardware alone or in a combination of hardware and software (for example programmable DSPs) and that the current invention should not be limited by ' the form of -the implementation.

Abstract

A non-intrusive test method for video sequences is described that correlates a number of quality indicators derived from a sequence with a matrix of constants to predict a quality score for the sequence.

Description

METHOD FOR TESTING VIDEO SEQUENCES
This invention relates to a test method that is can assess the quality of video sequences, in particular digitally encoded video signals. In order to reduce the bandwidth necessary to carry video sequences, compression algorithms such as H263 or MPEG rely on a set of well known methods. The first of these is to suppress the information present in each frame that is not perceptually relevant. The result of this operation may not be perceptible to the end user, perceptible but not perceptually relevant, or disruptive to the information content of the frame. In most cases, the effect of the information suppression will be to modify the appearance of textured objects, giving a greater repetitiveness or simplicity.
In real world applications, where the video quality has to be assessed continuously, it can be very difficult or costly to provide a system with a test signal, and the test signal will be transmitted through the system under test. For these reasons, non-intrusive systems have been investigated, where only a reconstructed signal is needed. How can we know the quality of a video sequence be assessed without being able to determine what has been added to or removed from it? However, as a subjective viewer knows what quality of video sequence can be expected from a television set, most people are able to make a judgement' of quality without referring to the original image. More advanced or experiences users of the system will also know the type of degradation introduced by the medium they use, and will be able to spot them to make a more educated estimation of the quality. Similarly, it is known that each family of compression algorithm tends to introduce its own type of distortion and what can be expected from natural images. This information can be used to determine what amount of distortion has been added to a sequence without referring to its original.
According to a first aspect of the invention there is provided a method of comparing two frames from a video sequence, the method comprising the steps of;
(i) taking a first frame from said video sequence; (ii) taking the subsequent frame from said sequence; (iii) comparing one pixel from the first frame with the corresponding pixel from the subsequent frame; and
(iv) repeating step (iii) for all pixels in a test area of the first frame.
If the results of the comparison stage indicate a large number of substantially equal pixels then a dropped frame may be inferred. If the results of the comparison stage indicate a small number of substantially equal pixels then a scene cut may be inferred.
According to a second aspect of the invention there is provided a method of predicting a quality score for a video sequence using quality parameters derived from a video sequence as described above
The invention will now be described, by way of example only, with reference to the following figures in which;
Figure 1 shows a reference video frame; Figure 2 shows the video frame of Figure 1 after having been compressed;
Figure 3 shows a schematic depiction of apparatus for performing a method according to the present invention;
Figure 4 shows the variation in unchanged pixel proportion with time for a number of different compression schemes; Figure 5 shows the correlation of predicted and subjective quality scores for a number of video sequences; and
Figure 6 shows the correlation of predicted and subjective quality scores for a number of subsets of video sequences
Most video sequences used come from natural sources, such as a studio or external scene captured by a video camera. This type of transceiver produces white noise, which implies that the value for a given pixel is likely to change over time even if the scenery being analysed is still and in a perfectly controlled environment. Because of its low amplitude and high spatial frequency, this noise is the first feature that will be modified or removed by compression algorithms. For this reason, absence of noise is an indicator of compression level. This effect is shown in the reference and degraded frames shown in Figures 1 and 2 respectively. The details present in the tarmac textures, such as the light presence of noise and the gradient have been simplified and replaced by a "flat" area of the average luminance and colour.
In most compression algorithms, this simplification operation is done within blocks of pixels, most often of 8x8 pixels. As a consequence, disparities tend to appear between adjacent regions, making the block structure of the image analysis visible. This effect, referred as blockiness, can easily be measured (see "Image quality measure based on local visual properties", P Franti, Picture Coding Symposium 97, 10-1 2 Sept. 1997, no.143, pp217-20, or M Ghanbari, "Blockiness Detection for MPEG2-coded video", IEEE Signal Processing Letters, 7:8 (August 2000), pp 21 3-21 5) even without having access to the reference sequence.
A second such method used to reduce bandwidth requirements for video transmissions is movement estimation. This technique involves analysing the motion present between one frame and the next frame in the sequence, so that movement vectors can be transmitted to the decoder in order to reconstruct the sequence from a starting frame. In order to avoid the accumulation of errors over time, motion estimation is not used between every frame. Instead, a temporal pattern can be used, as in the MPEG-2 compression scheme, where original, interpolated and predicted frames are interleaved in a predefined sequence. The third method used is to reduce the frame rate of the sequence. This is used for lower bitrate applications such as videoconferencing or video over IP, and is simply done by dropping one or more frames from a sequence.
It appears from these three main effects that motion present in a sequence is an important indicator of the image quality. Furthermore, the motion present in a sequence can be determined without the help of the original sequence, enabling the creation of non intrusive assessment systems that can estimate the quality of video systems.
Figure 3 shows video capture unit 10, which is suitable for capturing a series of frames from a video sequence. Each frame from the series is sent to blockiness. detector 40 and also each frame is transmitted to the comparator 20, along with the subsequent frame from the video sequence. Both the blockiness detector 40 and the comparator 20 are connected to a data processing unit 30, so that outputs from the blockiness detector and the comparator can be passed to the data processing unit.
The blockiness detector 40 analyses the pixels of each video frame received from the video capture unit. As has been described above the video compression technique divides the video image into a number of blocks and a side effect of simplifying the content of each block is to increase the disparity between the blocks, making them clearly visible in the reconstructed image. Different approaches have been tried to detect blockiness (see Franti or Ghanbari op cit) but the preferred technique for detecting blockiness for use in the present invention is to compare the variation of a pixel property (for example the luminance) every 8 pixels with the sum of variations within these 8 pixels. A ratio indicating a higher activity every 8 pixels than within 8 pixel blocks triggers an alarm as it tends to indicate an abrupt change in the pixel property in the transition from a block of very similar values to the next block. Alarms are counted across the whole frame, and the sum is used as an indicator of blockiness for the image. These blockiness indicator values are then transmitted to the data processing unit, along with a reference to the frame to which they correspond.
The two video frames are then sent to comparator 20 which compares the two frames pixel by pixel using one of the properties of a pixel as a basis for the comparison. The results of the comparator are fed to data processing unit 30. The comparator counts all those pixels where there has been no significant change in the pixel property (i.e. the luminance of the pixel in the subsequent frame is within a pre-determined threshold from the luminance of the pixel in the first frame) and determines the proportion of pixels for which the pixel property has not changed. The proportion of pixels in which there has been no change of pixel property and/or the variation in the proportion of pixels in which there has been no change of pixel property can be used to assess the quality of the video compression. Data processing unit 30 stores the pixel proportion values along with a reference to the frame to which they correspond. Most video capture systems introduce some noise into the image. For instance, the luminance of a given pixel produced by a video camera which is capturing a stationary image, for example background scenery, will vary in time. As the video compression rate is increased, i.e. the bandwidth of the transmitted video stream is decreased, the variations of a given pixel tend to be suppressed, and the pixel properties (for example chrominance, luminance, etc.) tend towards a constant value. A result of this is that stationary objects present in a compressed video sequence tend to look unnaturally still. This is shown in Figure 4, where Figure 4a shows the variation of the proportion of unchanged pixels for the reference signal, Figure 4b shows the variation of the proportion of unchanged pixels when that signal has been compressed at 4Mb/s and Figure 4c shows the variation of the proportion of unchanged pixels when the reference signal has been compressed at 1 Mb/s. Thus this variation can also be used to indicate the level of compression that has been applied to a reference signal. Accordingly the data processing unit has the facility to calculate a value which gives an indication of the variation of pixel proportion values# e.g. by performing a fast Fourier transform (FFT) of the pixel proportion values, calculating the derivative of the pixel proportion value, etc. The preferred indicator is to measure the average and standard deviation of the proportion over a 500m window.
If only a small number of pixels, for example less than 4%, have remained the same then it can be assumed that there has been a cut in the video sequence (i.e. the same scene is being viewed from a different camera angle, or a new scene is now being viewed). As it is not possible to proceed with the analysis of the change in pixel values for these frames because of the scene cut, the analysis may be discontinued and the next frame in the sequence may be selected for comparison.
If a very large number of pixels, for example greater than 98%, have remained the same then there is a high probability that a frame has been dropped, i.e. the same frame has been repeated due to the failure of the subsequent frame in the sequence to be received. The ratio of dropped frames is recorded by the data processing unit, and their position in the sequence.
If a large number of dropped frames are detected within a video sequence then this may be an indicator that the video sequence is being transmitted across an unreliable transport mechanism (in this case if a video sequence is being transmitted 'live' across a communications network, it may be possible to send a signal to the transmitting codec to decrease the frame rate in use (if the frame rate received is less than the frame rate in use) or to switch to a transport mechanism having a greater quality of service.
These three indicators, the amount of blockiness, the proportion of unchanging pixels and the ratio of dropped frames can be used to generate a model that can be used to assess the quality of a received video sequence in a non-intrusive manner, that is without needing to insert any test signals, or without having access to the initial video sequence. A matrix was formed from each of the three indicators and a time-dependent variation of each indicator. The non- intrusive model was trained and calibrated using linear-regression techniques to assess the qualities of a number of known video sequences which had previously been subjectively judged (that is, the quality of the sequences has been assessed by a number of human viewers).
The non-intrusive model, by design, is looking for known artefacts introduced by the codec in use. For this reason, the characteristics of the codec in use must be known and the model trained for the particular configuration being assessed. This requirement is incompatible with applications such as assessment of quality improvement of compression algorithms or tests on new coding schemes for instance. The model could easily be tricked by hiding such artefacts, without improving the overall quality (the most obvious example would be the use of a blockiness smother algorithm, as described by Ghanbari op cit). On the other hand, this approach can be particularly useful to telecom operators who are not changing coding algorithms on daily basis, but who want to monitor the performance of their network.
The first database used to calibrate and train the non-intrusive model is the series of sequences used for the VQEG contests. It consists of a number of sequences which are repeated but at different levels of compression. The resolution used in the sequences and the levels of degradation are consistent with broadcast applications (for the purpose of this calibration, some degradations are discarded, because they are based on analogue techniques. The second database relates to videoconference applications. The format used is either CIF or QCIF, the codec used to generate the degradations are H263 or MPEG4, and IP or UMTS error pattern have been inserted into the data stream before the decoding stage. The main difference between the two databases lies in the way the sequences have been previously assessed. The opinion scores of the VQEG sequences, which are about 10 seconds long in average, have been taken at the end of each presentation. The second database, which contains longer sequences and shows rapid variations in quality, has been assessed continuously. As test material is complicated to produce, the databases were split into a calibration set and a test set.
The test set of each database were processed through the model, which produces a set of parameters for each frame it contains (blockiness, pixel variation and dropped frame ratio). An extra set of parameters is derived from these initial data, by computing their mean and mean and standard variation on blocks of one second. These account for the temporal variations, which are linked to the compression rate. A linear regression is used to obtain the best combination between the extracted parameters and the mean opinion score as previously measured for the given sequences. The combination is then tried with on the parameters extracted for the rest of the database. The produced is compared with the measured mean opinion score, giving us an indicator of the model performance. Once the model linear regression technique has been applied a number of times a composite node is arrived at that can be used to predict the subjective testing of the sequences in the test set. The model was also calibrated using a subset of the Eurescom Aquavit database. This bank of subjectively assessed sequences represents a variety of codec and compression rate at levels compatible with multimedia applications and Figure 5 shows the correlation between the predicted score generated by the non- intrusive model and the mean subjective score for the test set of sequences. Figure 6a shows the variation of correlation between predicted score and subjective score with the frame rate used in the sequence and Figure 6b shows the variation of correlation between predicted score and subjective score with the bandwidth used to code the sequence.
It will be realised that another pixel property other than luminance, for example chrominance, hue or one of the RGB signals, could be used to assess the inter-frame variation of pixels. Additionally, all of the pixels in the frame could be compared with their respective pixels in a second frame or a specific sub-set of pixels could be compared with their respective pixels in the second frame. This organisation of the coded sequence has an effect on the distribution of the motion. In a non coded sequence, the amount of inter-frame change will change slowly in relation to the variation of motion present in the image. However, in the coded sequence, the inter-frame change varies quickly around its average value, with a period related to the coding sequence used by the encoder.
Profiles can be incorporated into the model, according to the type of video sequences processed. Requirements and expectations can be very different according to the type of service or images transmitted. At the equivalent level of quality, a video sequence can be acceptable in video conference but not in broadcast. The bandwidth requirement are much higher for sport than head and shoulder sequences. Profiles take these aspects into account before to make a judgement about the quality.
The measurement tool can be used in different ways. By providing a means to measure the quality of refreshment in video sequences, it allows comparison between a reference signal and the same signal once gone through by the coding and decoding process. But it can also provides useful information about a sequence without the help of any reference.
If processing power is limited (for example if a real-time assessment is required or the model is being incorporated into a device with limited capabilities then a less-sophisticated model can be derived by using fewer parameters; the nature of the system to be measured will influence which quality indicator, or its time dependence, can be excluded from the model and the optimal combination can easily be determined by experimentation and further application of linear regression techniques. It will be readily understood that the present invention could be implemented in software alone, i.e. running on a general purpose computer such as personal computer (which may be supplied on a data carrier such as, for example a floppy disc of a CD-ROM), in hardware alone or in a combination of hardware and software (for example programmable DSPs) and that the current invention should not be limited by 'the form of -the implementation.

Claims

1 . A method of comparing two frames from a video sequence, the method comprising the steps of; (i) taking a first frame from said video sequence;
(ii) taking the subsequent frame from said sequence; (iii) comparing one pixel from the first frame with the corresponding pixel from the subsequent frame; and
(iv) repeating step (iii) for all pixels in a test area of the first frame.
2. A method according to claim 1 wherein the luminance of the corresponding pixels is compared.
3. A method according to claim 1 wherein if the results of the comparison stage indicate a large number of substantially equal pixels then a dropped frame is inferred.
4. A method according to claim 1 wherein if the results of the comparison stage indicate a small number of substantially equal pixels then a scene cut is inferred.
4. A method of assessing the quality of a video sequence, the method comprising the steps of
(i) comparing all of the frames in the video sequence with the subsequent frame in accordance with claim 1
(ii) ignoring those comparisons indicating a dropped frame or a scene cut
(iii) calculating a quality figure from the remaining comparison values
PCT/GB2001/002374 2000-05-26 2001-05-29 Method for testing video sequences WO2001091473A1 (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
US10/275,474 US7233348B2 (en) 2000-05-26 2001-05-29 Test method
CA2408435A CA2408435C (en) 2000-05-26 2001-05-29 Method for testing video sequences
EP01931970A EP1290900A1 (en) 2000-05-26 2001-05-29 Method for testing video sequences
AU58652/01A AU5865201A (en) 2000-05-26 2001-05-29 Method for testing video sequences

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
GB0012992.4 2000-05-26
GBGB0012992.4A GB0012992D0 (en) 2000-05-26 2000-05-26 Test method

Publications (1)

Publication Number Publication Date
WO2001091473A1 true WO2001091473A1 (en) 2001-11-29

Family

ID=9892529

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/GB2001/002374 WO2001091473A1 (en) 2000-05-26 2001-05-29 Method for testing video sequences

Country Status (6)

Country Link
US (1) US7233348B2 (en)
EP (1) EP1290900A1 (en)
AU (1) AU5865201A (en)
CA (1) CA2408435C (en)
GB (1) GB0012992D0 (en)
WO (1) WO2001091473A1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2006129059A1 (en) * 2005-06-02 2006-12-07 British Telecommunications Public Limited Company Video signal loss detection
WO2008077160A1 (en) * 2006-12-22 2008-07-03 Mobilkom Austria Aktiengesellschaft Method and system for video quality estimation

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090309977A1 (en) * 2008-06-12 2009-12-17 Microsoft Corporation Benchmarking and calibrating video quality assessment tools
US8294772B2 (en) * 2009-01-29 2012-10-23 Pelco, Inc. System and method for monitoring connections within an analog video system
EP2647199B1 (en) * 2010-11-30 2017-01-11 Thomson Licensing Method and apparatus for measuring quality of video based on frame loss pattern
CN103839263B (en) * 2014-02-26 2016-07-06 宁波大学 A kind of stereo-picture depth perception method for objectively evaluating
US10979744B2 (en) 2017-11-03 2021-04-13 Nvidia Corporation Method and system for low latency high frame rate streaming

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5589884A (en) * 1993-10-01 1996-12-31 Toko Kabushiki Kaisha Adaptive quantization controlled by scene change detection
EP0938239A1 (en) * 1998-02-20 1999-08-25 Tektronix, Inc. Low duty-cycle transport of video reference images
US6014183A (en) * 1997-08-06 2000-01-11 Imagine Products, Inc. Method and apparatus for detecting scene changes in a digital video stream

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5446492A (en) * 1993-01-19 1995-08-29 Wolf; Stephen Perception-based video quality measurement system
US5627765A (en) * 1994-07-25 1997-05-06 Avid Technology, Inc. Method and apparatus for compressing and analyzing video and for creating a reference video
US5754700A (en) * 1995-06-09 1998-05-19 Intel Corporation Method and apparatus for improving the quality of images for non-real time sensitive applications
DE19521408C1 (en) 1995-06-13 1996-12-12 Inst Rundfunktechnik Gmbh Objective evaluation of two or three dimensional pictures
GB9604315D0 (en) * 1996-02-29 1996-05-01 British Telecomm Training process
US6119083A (en) * 1996-02-29 2000-09-12 British Telecommunications Public Limited Company Training process for the classification of a perceptual signal
US5767922A (en) * 1996-04-05 1998-06-16 Cornell Research Foundation, Inc. Apparatus and process for detecting scene breaks in a sequence of video frames
US6539055B1 (en) * 1999-12-03 2003-03-25 Intel Corporation Scene change detector for video data

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5589884A (en) * 1993-10-01 1996-12-31 Toko Kabushiki Kaisha Adaptive quantization controlled by scene change detection
US6014183A (en) * 1997-08-06 2000-01-11 Imagine Products, Inc. Method and apparatus for detecting scene changes in a digital video stream
EP0938239A1 (en) * 1998-02-20 1999-08-25 Tektronix, Inc. Low duty-cycle transport of video reference images

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2006129059A1 (en) * 2005-06-02 2006-12-07 British Telecommunications Public Limited Company Video signal loss detection
US8228386B2 (en) 2005-06-02 2012-07-24 British Telecommunications Public Limited Company Video signal loss detection
WO2008077160A1 (en) * 2006-12-22 2008-07-03 Mobilkom Austria Aktiengesellschaft Method and system for video quality estimation

Also Published As

Publication number Publication date
CA2408435A1 (en) 2001-11-29
GB0012992D0 (en) 2000-07-19
EP1290900A1 (en) 2003-03-12
US7233348B2 (en) 2007-06-19
US20030142214A1 (en) 2003-07-31
AU5865201A (en) 2001-12-03
CA2408435C (en) 2010-11-02

Similar Documents

Publication Publication Date Title
US9037743B2 (en) Methods and apparatus for providing a presentation quality signal
Vranješ et al. Review of objective video quality metrics and performance comparison using different databases
US9143776B2 (en) No-reference video/image quality measurement with compressed domain features
Leszczuk et al. Recent developments in visual quality monitoring by key performance indicators
Winkler Video quality and beyond
JP2016530751A (en) A concept for determining the quality of media data streams with varying quality versus bit rate
WO2012013777A2 (en) Method and apparatus for assessing the quality of a video signal during encoding or compressing of the video signal
Reibman et al. Predicting packet-loss visibility using scene characteristics
Huynh-Thu et al. Impact of jitter and jerkiness on perceived video quality
Huynh-Thu et al. No-reference temporal quality metric for video impaired by frame freezing artefacts
CA2408435C (en) Method for testing video sequences
Leszczuk et al. Key indicators for monitoring of audiovisual quality
WO2008077160A1 (en) Method and system for video quality estimation
KR20100071820A (en) Method and apparatus for measuring quality of video
Shanableh No-reference PSNR identification of MPEG video using spectral regression and reduced model polynomial networks
CA3168392A1 (en) Real-time latency measurement of video streams
Nur Yilmaz A no reference depth perception assessment metric for 3D video
Brandas et al. Quality assessment and error concealment for SVC transmission over unreliable channels
Wang et al. Network-based model for video packet importance considering both compression artifacts and packet losses
Punchihewa et al. A survey of coded image and video quality assessment
KR101083063B1 (en) Method and apparatus for measuring video quality of experience
Ong et al. Video quality monitoring of streamed videos
Soares et al. No-reference lightweight estimation of 3D video objective quality
Miličević et al. An approach to interactive multimedia systems through subjective video quality assessment in H. 264/AVC standard
Hands et al. Video QoS enhancement using perceptual quality metrics

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 EC 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: 10275474

Country of ref document: US

WWE Wipo information: entry into national phase

Ref document number: 2408435

Country of ref document: CA

WWE Wipo information: entry into national phase

Ref document number: 2001931970

Country of ref document: EP

WWP Wipo information: published in national office

Ref document number: 2001931970

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: JP