« PrécédentContinuer »
United States Patent  [li] Patent Number: 4,682,230
Perlman et al.  Date of Patent: Jul. 21, 1987
U.S. Patent Jul. 21, 1987 Sheet 1 of 2 4,682,230
 ADAPTIVE MEDIAN FILTER SYSTEM
 Inventors: Stuart S. Perlman, Princeton
Township, Mercer County; Sanford
Eisenhandler, East Windsor; Paul W.
Lyons, Plumstead Township, Ocean
County; Michael J. Shumila,
Hamilton Township, Mercer
County, all of N.J.
 Assignee: RCA Corporation, Princeton, N.J.
 Appl. No.: 842,644
 Filed: Mar. 21,1986
 Int. CI." H04N 5/21
 U.S. CI 358/167; 358/36;
 Field of Search 358/167, 36, 166, 37,
358/905, 222, 168, 169; 455/296, 302, 303, 307
 References Cited
U.S. PATENT DOCUMENTS
4,177,430 12/1979 Paul 455/307
4,398,296 8/1983 Gott 375/103
4,402,013 8/1983 Wargo 358/160
4,464,686 8/1984 Reitmeier 358/36
4,485,399'11/1984 Schulz 358/167
4,517,600 5/1985 Reitmeier 358/167
4,562,470 12/1985 Dinh 358/167
"Handbook of Tables for Probability and Statistics", edited by W. H. Beyer, 1966, p. 2. I. Scollar et al.; "Image Enhancement Using the Median and the Interquartile Distance"; Computer Vision, Graphics, and Image Processing; Academic Press, Inc.; 1984; pp. 236-251.
"VLSI Architecture for a One-Chip Video Median Filter"; Demassieux, et al.; Proceedings IEEE International Conference on Acoustics, Speech and Signal Processing, vol. 3, pp. 1001-1004 (1985).
Primary Examiner—Tommy P. Chin
Attorney, Agent, or Firm—Paul J. Rasmussen; Eric P.
Herrmann; David N. Caracappa
An adaptive median filter system is disclosed. Circuitry is arranged to produce successive sets of samples from an input signal which may possibly include noise. An adaptive median filter filters the samples in response to a control signal. Further circuitry estimates the relative density of the noise in the input signal to generate the control signal supplied to the adaptive median filter.
8 Claims, 4 Drawing Figures
U.S. Patent Jul. 21, 1987 Sheet 2 of 2 4,682,230
ADAPTIVE MEDIAN FILTER SYSTEM
The present invention relates to adaptive median filter systems. 5
Developments in image processing have made possible the processing images in a variety of forms. For example, a photograph may be processed as a two-dimensional array of samples. A video signal, as from a television camera, may be processed as a three-dimen- 10 sional array of samples —the third dimension being the temporal dimension and representing motion in the image. One filtering algorithm which may be applied in the processing of such images is the median filter.
A median filter reduces impulse-type noise in any 15 signal but may be particularly efficacious for reducing such noise in an image such as may be caused by a scratch in a photograph or a dropout in the communications channel over which the video signal is transmitted. A median filter does not adversely affect edges or cor- 20 ners which exist in an image, however. In prior art median filters, the current sample (the sample being filtered) was replaced by a sample having the median value of the values of that sample and a number of samples surrounding that sample.
An enhancement of the median filtering algorithm described above selectively substitutes the sample having the median value for the current sample only if the value of the current sample lies outside of predeter- 3Q mined thresholds. For example, in the paper "Image Enhancement Using the Median and Interquartile Distance" by Scollar et al. in Computervision, Graphics and Image Processing, Vol. 25, pages 236-251 (1984), the difference between the value of the current sample 35 and the median value is compared to a constant times the difference between the upper and lower quartiles (the interquartile distance). If the current sample/median difference is greater than the interquartile distance function, then the sample having the median value 40 is substituted for the current sample. Scollar et al. have found this algorithm advantageous because it "adapts to local contrast in the data, and sharp edges will hardly be blurred at all, . . . ".
In accordance with the principles of the present in- 45 vention, a further improvement may be achieived if non-image related criteria are used for adapting the median filter algorithm, in addition to the contrast related criteria described in the above Scollar paper. Because one of the primary advantages of a median filter is 50 its ability to attenuate impulse noise, improved operation may be attained if the operation of the adaptive median filter is modified in response to the relative density of impulse noise in the signal being filtered.
Apparatus in accordance with the principles of the 55 present invention comprises a source of an input signal possibly including noise, and a means for producing a plurality of samples representing the signal. An adaptive median filter filters these samples in response to a control signal. A noise estimator estimates the relative 60 density of the noise in the signal being sampled. A control signal generator generates the control signal supplied to the adaptive median filter in response to the estimated noise density.
In the drawings: 65
FIG. 1 is a block diagram of an adaptive median filter system in accordance with the principles of the present invention;
FIG. 2 is a block diagram of an adaptive median filter which may be used in the system illustrated in FIG. 1;
FIG. 3 is a block diagram of an M-tile producer which may be used in the median filter illustrated in FIG. 2; and
FIG. 4 is a block diagram of an alternative embodiment of an adaptive median filter system in accordance with the principles of the present invention.
In FIG. 1, the input signal being median filtered, for example, a video signal from the front end of a television receiver (not shown) is applied to an input terminal 5. Input terminal 5 is coupled to a sample producer 10 and a noise estimator 30. Sample producer 10 produces a plurality of samples, including a current sample (the sample currently being filtered) representing a given point in the image being processed, and other samples representing points surrounding that point. These samples are coupled to respective data input terminals of an adaptive median filter 20. Adaptive median filter 20 produces at an output terminal a median filtered output signal. An output terminal of noise estimator 30 is coupled to an input terminal of a control signal generator 40 which generates a control signal at an output terminal which is coupled to a control input terminal of the adaptive median filter 20.
In operation, sample producer 10 produces versions of the input signal from terminal 5 which represent the current and surrounding points. In a system in which the input signal is a video signal, sample producer 10 may comprise a tapped delay line which provides a current sample and samples surrounding the current sample either vertically, horizontally, temporally or some combination of these. The adaptive median filter 20 selectively couples either the current sample or a sample having the median value of the current and surrounding samples to the filter output terminal. The decision to substitute a median value sample for the current sample is based upon the values of the input samples and the control signal supplied to the adaptive median filter 20.
Noise estimator 30 evaluates the input signal supplied at input terminal 5 to determine the relative density of the noise in the input signal. Because the adaptive median filter is particularly useful for attenuating impulse noise, noise estimator 30 may, for example, detect the relative density of impulse noise in the input signal. Noise estimator 30 may comprise a threshold detector which produces a pulse whenever the input signal exceeds the thresholds, indicating a noise impulse; and a resettable counter which counts the number of pulses over a predetermined interval of time.
The control signal is generated by control signal generator 40 responsive to the density of impulse noise to more completely attenuate the impulse noise without adversely affecting the intelligence transmitted in the signal. Contol signal generator 40 may, for example, be a decoder for generating control signals having different values which correspond to different ranges of estimated noise density. In a digital implementation, this decoder may e preprogrammed digital memory.
FIG. 2 illustrates an adaptive median filter 20 which may be used in the apparatus illustrated in FIG. 1. In FIG. 2, samples from sample producer 10 of FIG. 1 are coupled to data input terminals of an M-tile producer 21. The control signal from control signal generator 40 is coupled to a control input terminal of M-tile producer