Adaptive spatial-temporal postprocessing for low bit-rate coded image sequences

• US 5,512,956 A
• Filed: 02/04/1994
• Issued: 04/30/1996
• Est. Priority Date: 02/04/1994
• Status: Expired due to Term

• Current Assignee: American Telephone & Telegraph Company (AT&T, Inc.)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Yan, Li
• Primary Examiner(s): Groody, James J.
• Assistant Examiner(s): Murrell, Jeffrey S.
• Application Number:
  US08/191,685
• Time in Patent Office: 816 Days
• Field of Search: 348/412, 348/413, 348/421, 348/415, 348/416, 348/606, 348/607, 348/620, 348/627, 348/619, 348/497
• US Class Current: 348/606
• CPC Class Codes: H04N 19/527 Global motion vector estimation
• CPC Class Codes: H04N 19/86 involving reduction of coding artifacts, e.g. of blockiness

Abstract

A 3D nonlinear postprocessing system and method are utilized to reduce coding artifacts produced by block-based motion-compensated transform coding. In the system and method, a separable 3D filtering structure is used: space-variant FIR-Median Hybrid filtering is used in spatial domain, followed by a motion-compensated nonlinear filtering in the temporal domain. By using this structure and method, the coding artifacts in a reconstructed image sequence can be effectively reduced without blurring edges or moving objects in the image sequence. Significant improvement in the picture quality of low bit-rate coded video sequences is thereby achieved.

Claims

1. A filtering system for reducing artifacts in motion video sequences generated by block-based motion-compensated transform coding from a video decoder, comprising:

a postprocessor connected to said video decoder, said postprocessor including;

an edge orientation detector;

a spatial filter bank, said spatial filter bank being configured to receive information from said edge orientation detector, said filter bank comprising a one-dimensional filter utilizing said information to generate spatially filtered video sequences; and

a motion-compensated temporal filter, said motion-compensated temporal filter receiving spatially filtered video sequences generated by said spatial filter bank, said motion-compensated temporal filter being configured to generate temporally filtered video sequences from said spatially filtered video sequences; and

a frame memory in communication with said spatial filter bank, said frame memory being arranged to receive spatially filtered video sequences from said spatial filter bank.

2. The filtering system of claim 1 wherein said edge orientation detector utilizes a set of template gradient impulse response arrays to compute edge orientation.

3. The filtering system of claim 2 wherein said edge orientation detector:

convolves an image with said gradient impulse response array;

determines a maximum gradient; and

compares said maximum gradient to a threshold to determine whether a pixel belongs to an edge class or a nonedge class.

4. The filtering system of claim 1 wherein said spatial filter bank comprises at least two finite impulse response linear phase filters and a median filter.

5. The filtering system of claim 4 wherein said spatial filter bank comprises:

an index to indicate a direction of operation for said finite impulse response linear phase filters, and

a central input pixel from which said finite impulse response filters are measured.

6. The filtering system of claim 1 wherein said spatial filter bank comprises a finite impulse response median filter bank having varying root structures for filtering pixels in said motion video sequence.

7. The filtering system of claim 1 wherein said spatial filter bank further comprises a two dimensional low-pass filter for filtering nonedge pixels in said motion video sequence.

8. The filtering system of claim 1 wherein said frame memory is configured to store signals for a current frame, an adjacent prior frame and an adjacent future frame.

9. The filtering system of claim 8 wherein said motion-compensated temporal filter is configured to:

determine a best matching frame from said adjacent prior frame and a motion-compensated frame; and

calculate a frame difference between a current frame and said best matching frame.

10. The filter system of claim 9 wherein said motion-compensated temporal filter is configured to determine said best matching frame from best matching frame pixels selected from either said adjacent prior frame or said motion-compensated frame, said best matching frame pixels comprising:

pixels from said adjacent prior frame when the absolute difference between pixels in said current frame and pixels in said motion-compensated frame is greater than the absolute difference between pixels in said current frame and said adjacent prior frame; and

pixels from said motion-compensated frame when the absolute difference between pixels in said current frame and pixels from said adjacent prior frame is greater than the absolute difference between pixels in said current frame and pixels in said motion-compensated frame.

11. The filter system of claim 9 wherein said motion-compensated temporal filter further comprises a nonlinear mapper to determine a weighting factor based, at least in part, on the first frame difference.

12. The filtering system of claim 11 wherein said motion-compensated temporal filter is configured to calculate a weighted average current frame from said current frame and said best matching frame based, at least in part, on the weighting factor.

13. The filtering system of claim 12 wherein said motion-compensated temporal filter is configured to:

calculate a second frame difference between said weighted average current frame with a future frame to generate a second frame difference;

determine a second weighting factor based, at least in part, on the second frame difference; and

calculate a second weighted average of the weighted average current frame and the adjacent future frame based, at least in part, on the second weighting factor.

14. A system for filtering decoded noise-contaminated signals for video communication, comprising:

a computer processor;

means for filtering spatial artifacts from decoded noise-contaminated signals, said means for filtering spatial artifacts comprising an edge detector and a one-dimensional filter;

means for storing signals from said decoded noise-contaminated signals for at least a current frame, a prior frame and a future frame;

means for calculating a best matching frame from said prior frame and a motion-compensated frame;

means for calculating an intermediate weighted average frame from said current frame and a best matching frame such that the motion-compensated frame is given less weight as the difference between said intermediate weighted average frame and said motion-compensated frame increases; and

means for calculating a filtered frame from a weighted average of the intermediate weighted average current frame and said future frame, such that said future frame is given less weight as the difference between said intermediate weighted average current frame and said future frame increases.

15. The system of claim 14 wherein said means for filtering spatial artifacts utilizes a set of template gradient impulse response arrays to compute edge orientation.

16. The system of claim 15 wherein said means for filtering further comprises means for convolving an image with said gradient impulse response array, means for determining a maximum gradient, and means for comparing said maximum gradient to a threshold to determine whether a pixel belongs to an edge class or a nonedge class.

17. The system of claim 16 wherein said means for filtering spatial artifacts comprises at least two finite impulse response linear phase filters and a median filter.

18. The system of claim 17 wherein said means for filtering spatial artifacts further comprises a two-dimensional filter, wherein said two-dimensional filter filters spatial artifacts from pixels belonging to said nonedge class, and wherein said one-dimensional filter filters spatial artifacts from pixels belonging to said edge class.

19. A three dimensional filter system for enhancing decompressed motion video sequences generated by block-based motion-compensated transform coding, comprising:

a one-dimensional space-variant FIR-median hybrid filter, said one-dimensional space-variant FIR-median hybrid filter being structured to reduce the effect of spatial artifacts generated by block-based motion-compensated transform coding on said motion video sequences;

a memory to store least a portion of said motion video sequences, said memory being connected to said one-dimensional space-variant FIR-median hybrid filter; and

a motion-compensated nonlinear filter, said motion-compensated nonlinear filter being structured to reduce the effect of temporal artifacts generated by block-based motion-compensated transform coding on said motion video sequences;

wherein said one-dimensional space-variant FIR-median hybrid filter, said memory and said motion-compensated nonlinear filter are arranged in series.

20. The filter system of claim 19 further comprising:

a one-dimensional edge detector, said edge detector being configured to input information to said space-variant FIR-median hybrid filter to reduce the effect of spatial artifacts.

21. A method of processing and displaying decoded video image sequence signals generated by block-based motion-compensated transform coding, comprising the steps of:

detecting the edge orientation of a pixel of said video image sequences;

filtering spatial artifacts from said video image sequences using a one-dimensional filter for pixels determined to have an orientation in an edge region of said video sequence signals;

storing at least a current frame, an adjacent prior frame and an adjacent future frame from said video image sequence signals;

calculating a temporally filtered frame from a weighted average of said adjacent future frame, said current frame and a motion-compensated adjacent prior frame; and

displaying said temporally filtered frame.

22. The method of claim 21 wherein said step of calculating a temporally filtered frame further comprises the steps of:

calculating a best matching frame;

calculating a frame difference between said current frame and said best matching frame;

determining a weighting factor based, at least in part, on said frame difference;

calculating a weighted average current frame from the current frame and said best matching frame, based, at least in part, on the weighting factor.

23. The method of claim 22 wherein said step of calculating a temporally filtered frame further comprises the steps of:

calculating a second frame difference between said weighted average current frame and said adjacent future frame,

determining a second weighting factor based, at least in part, on the second frame difference, and

calculating a weighted average of the weighted average current frame and said adjacent future frame based, at least in part, on said second weighting factor.

24. The method of claim 21 wherein said step of detecting the edge orientation of a pixel comprises the steps of:

convolving said pixel with a gradient impulse response array;

determining a maximum gradient; and

comparing said maximum gradient to a threshold to determine whether a pixel belongs in an edge region or a nonedge region.

25. The method of claim 21 wherein said step of filtering spatial artifacts from said video image sequences comprises the step of:

filtering spatial artifacts from said video image sequences using a two-dimensional low-pass filter for pixels determined to have an orientation in a nonedge region of said video sequences.

26. The method of claim 25 wherein said step of filtering spatial artifacts from said video image sequences comprises the steps of:

calculating the output of said two-dimensional low-pass filter as a function of a center pixel p(x,y) and all pixels from p(x-L, y-L) to p(x+L, y+L) where L is equal to (window length-1)/2.

27. The method of claim 21 wherein said step of filtering spatial artifacts further comprises the steps of:

selecting a direction from a pixel at the edge region p(c+L) through a center pixel p(c) to a second pixel at the edge region p(c-L);

calculating the output of a first finite impulse response filter as a function of a center pixel p(c) and all pixels from p(c) to p(c+L) along said direction;

calculating the output of a second finite impulse response filter as a function of center pixel p(c); and

calculating the output of a third finite impulse response filter as a function of all pixels from p(c-1) to p(c-L) along said direction;

wherein L is equal to (window length-1)/2.

28. The method of claim 27 wherein said step of filtering spatial artifacts further comprises the step of:

calculating the median of outputs of said finite impulse response filters.

29. The method of claim 28 wherein said step of filtering spatial artifacts further comprises the steps of:

selecting a direction of operation for said finite impulse response filters; and

calculating the overall output taken over the median of outputs of said finite impulse response filters in said direction of operation from a central input pixel from which said finite impulse response filters are measured.