Segmentation-based JPEG image artifacts reduction

• US 5,495,538 A
• Filed: 03/08/1995
• Issued: 02/27/1996
• Est. Priority Date: 05/28/1993
• Status: Expired due to Term

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Fan, Zhigang
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): Tran, Phuoc
• Application Number:
  US08/401,180
• Time in Patent Office: 356 Days
• Field of Search: 382/56, 382/54, 382/232, 382/233, 382/250, 382/251, 382/254260, 382/262, 382/264, 382/268, 358/432, 358/433, 358/447, 358/429
• US Class Current: 382/233
• CPC Class Codes: H04N 19/124 Quantisation
• CPC Class Codes: H04N 19/126 Details of normalisation or weighting functions, e.g. normalisation matrices or variable uniform quantisers
• CPC Class Codes: H04N 19/20 using video object coding
• CPC Class Codes: H04N 19/60 using transform coding
• CPC Class Codes: H04N 19/80 Details of filtering operations specially adapted for video compression, e.g. for pixel interpolation H04N19/635, H04N19/86 take precedence
• CPC Class Codes: H04N 19/86 involving reduction of coding artifacts, e.g. of blockiness

Abstract

A method for reducing ringing and blocking artifacts in a decompressed image models an image in a relatively small area as several smooth regions separated by edges. The method uses JPEG MxM pixel blocks and is compatible with JPEG decompression. To reduce ringing, a block is examined for uniformity, segmented and smoothed. Then, after a DCT transform, a projection is performed to guarantee that the DCT coefficients of the resulting image block will be within the initial quantization interval. The resultant image is produced by an inverse DCT. To reduce blocking, the method is modified to employ a large outer window for uniformity checking, segmentation and smoothing and a small inner window for DCT projection.

Claims

1. A method of decompressing a compressed image which reduces artifacts in a decompressed image, comprising the steps of:

inputting a signal representing a plurality of discrete cosine transform coefficients grouped into blocks, each of the plurality of input discrete cosine transform coefficients representing a corresponding pixel value for a pixel in the compressed image;

determining an initial quantization interval corresponding to each of the plurality of input discrete cosine transform coefficients based on the input discrete cosine transform coefficient;

performing an inverse discrete cosine transform on each block of input discrete cosine transform coefficients to produce a pixel block of a corresponding plurality of pixel values;

examining each pixel block for uniformity in pixel values;

grouping pixels having pixel values within an intensity threshold in each non-uniform pixel block into regions;

smoothing the pixel values for the pixels in each region by low pass filtering;

performing a discrete cosine transform on the smoothed pixel values for the pixels in each block, the discrete cosine transform yielding resultant discrete cosine transform coefficients;

projecting each of the resultant discrete cosine transform coefficients so that the resultant discrete cosine transform coefficients are within the corresponding initial quantization interval;

performing an inverse discrete cosine transform on the resultant discrete cosine transform coefficients to produce a final signal representing final pixel values for pixels corresponding to the decompressed image; and

outputting the final signal.

2. The method of claim 1, wherein the step of examining each pixel block for uniformity comprises comparing a range of pixel values for the pixels in each block to a uniformity threshold.

3. The method of claim 1, wherein the step of grouping pixels into regions comprises the steps of:

forming initial regions of connected pixels wherein neighboring pixels have a difference in pixel value smaller than the intensity threshold, and

merging neighboring initial regions into a single region in sequence, the merged neighboring initial regions having at most a minimum difference in mean pixel value.

4. The method of claim 1, wherein the step of projecting includes comparing each of the resultant discrete cosine transform coefficients for each block to the corresponding initial quantization interval and correcting the resultant discrete transform coefficients to fit within the initial quantization interval.

5. The method of claim 4, wherein, when the resultant discrete cosine transform coefficients for a block are not within the initial quantization interval, the step of correcting comprises partitioning at least one region of pixels within the block into two regions and repeating the steps of smoothing, transforming and comparing.

6. The method of claim 5, wherein the step of partitioning comprises splitting the last merged region first.

7. The method of claim 4, wherein when the discrete cosine transform coefficients fit generally within the initial quantization interval or a number of merged regions equals an established minimum number of regions, the method proceeds to the step of performing the inverse discrete cosine transform.

8. The method of claim 1, wherein the step of smoothing comprises replacing the pixel values of pixels in an interior of a region with a mean pixel value of pixels in the region, and replacing the pixel value of each pixel in the boundary of the region with a combination of the pixel value of the pixel and the mean pixel value of pixels in the region.

9. The method of claim 1, wherein the step of smoothing further comprises replacing the pixel values of boundary pixels with a pixel value obtained by the following relationship:

space="preserve" listing-type="equation">y(m,n)=α

[x]y(m,n)+(1-α

)[x]μ

where y(m,n) is the pixel value of a boundary pixel at position (m,n);

α

is a pre-set factor satisfying 0≦

α

≦

1; and

α

is the mean pixel value of the region.

10. The method of claim 1, wherein the step of projecting is performed by the following relationships:

space="preserve" listing-type="equation">Z(u,v)=U(u,v) if Y(u,v)>

U(u,v);

space="preserve" listing-type="equation">Z(u,v)=L(u,v) if Y(u,v)<

L(u,v); and

space="preserve" listing-type="equation">Z(u,v)=Y(u,v) otherwise,

wherein Z(u,v) is the output of the projection for (u,v) discrete cosine transform coefficients;

U(u,v) is an upper discrete cosine transform boundary;

L(u,v) is a lower discrete cosine transform boundary; and

Y(u,v) is a desirable discrete cosine transform boundary, wherein the upper and lower boundaries are based on the initial quantization interval.

11. The method of claim 1, further comprising:

forming at least one first outer window of pixel values around a corresponding second inner window of pixel values;

using the at least one first outer window for the steps of examining each pixel block for uniformity, forming regions, and smoothing; and

using the corresponding second inner window for the steps of performing the discrete cosine transform, projecting and performing the inverse cosine transform.

12. A method of decompressing compressed image data, comprising the steps of:

inputting a signal comprising blocks of discrete cosine transform coefficients, the discrete cosine transform coefficients representative of a compressed image;

establishing discrete cosine transform constraints based on the input discrete cosine transform coefficients;

transforming the blocks of discrete cosine transform coefficients into pixel blocks of intensity values representing pixels;

determining initial pixel intensity values for each of the pixel blocks;

determining uniformity of intensity values for pixels for each pixel block;

forming regions of pixels in each non-uniform pixel block;

replacing the initial pixel intensity values corresponding to pixels in a region with an average pixel intensity value of the respective region to smooth the region;

performing a discrete cosine transform on the intensity values for each pixel block yielding resultant discrete cosine transform coefficients;

projecting the resultant discrete cosine transform coefficients for each pixel block onto the established discrete cosine transform constraints;

performing an inverse discrete cosine transform on the projected resultant coefficients to produce a final signal comprising intensity values for pixels representative of a decompressed image; and

outputting the final signal.

13. The method of claim 12, wherein the step of determining uniformity comprises comparing an intensity range of values for the pixels in each block to a pre-set uniformity value threshold.

14. The method of claim 12, wherein the step of projecting comprises comparing the resultant coefficients to the established discrete cosine transform constraints.

15. The method of claim 14, wherein the step of projecting further comprises correcting the resultant discrete cosine transform coefficients to fit within the established constraints.

16. The method of claim 14, wherein when the resultant discrete cosine transform coefficients for a pixel block are not within the established coefficient constraints, the method further comprises partitioning a region in the pixel block into two and repeating the steps of replacing, transforming, and projecting.

17. The method of claim 12, wherein the step of projecting is performed by the following relationships:

space="preserve" listing-type="equation">Z(u,v)=U(u,v) if Y(u,v)>

U(u,v);

space="preserve" listing-type="equation">Z(u,v)=L(u,v) if Y(u,v)<

L(u,v); and

space="preserve" listing-type="equation">Z(u,v)=Y(u,v) otherwise,

wherein Z(u,v) is the output of the projection for (u,v) discrete cosine transform coefficients;

U(u,v) is an upper discrete cosine transform boundary;

L(u,v) is a lower discrete cosine transform boundary; and

Y(u,v) is a desirable discrete cosine transform boundary, wherein the upper and lower boundaries are based on the established discrete cosine transform constraints.

18. The method of claim 12, further comprising:

forming at least one first outer window of pixel intensity values around a corresponding second inner window of pixel intensity values;

using the at least one first outer window for the steps of determining uniformity, forming regions, and replacing pixel values; and

using the corresponding second inner window for the steps of performing the discrete cosine transform and projecting.

19. The method of claim 12, wherein the step of forming regions comprising the steps of:

forming initial regions of connected pixels wherein neighboring pixels have a difference in intensity value that is smaller than a pre-set intensity value threshold, and

merging neighboring initial regions into a single region in sequence, the merged neighboring initial regions having a minimum difference in mean pixel intensity value.

20. The method of claim 12, wherein the step of replacing pixel intensity values comprises replacing the initial pixel intensity value of each pixel in an interior of a region with a mean pixel intensity value of pixels in the region, and replacing the initial pixel intensity value of pixels in the boundary of the region with a combination of the initial pixel intensity value of the pixel and the mean pixel intensity value of pixels in the region.

21. The method of claim 12, wherein the step of replacing pixel intensity values further comprises replacing the initial pixel intensity value of boundary pixels with a value obtained by the following relationship:

space="preserve" listing-type="equation">y(m,n)=α

[x]y(m,n)+(1-α

)[x]μ

where y(m,n) is the intensity value of a boundary pixel at position (m,n);

α

is a pre-set factor satisfying 0≦

α

≦

1; and

μ

is the mean pixel intensity value of the region.