Lossy data compression exploiting distortion side information

• US 20060170571A1
• Filed: 12/07/2005
• Published: 08/03/2006
• Est. Priority Date: 12/09/2004
• Status: Active Grant

• Current Assignee: Massachusetts Institute of Technology
• Original Assignee: Massachusetts Institute of Technology
• Inventors: Zamir, Ram, Martinian, Emin, Wornell, Gregory W.
• Granted Patent: US 7,327,287 B2
• Time in Patent Office: Days
• US Class Current: 341/50
• CPC Class Codes: H03M 7/30 Compression speech analysis-synthesis for redundancy reduction G10L19/00; for image communication H04N; Expansion; Suppression of unnecessary data, e.g. redundancy reduction

Abstract

Described are techniques for performing lossy encoding. Source data and quality data are received by an encoder. The encoder maps the source data into a compressed representation having a level of distortion in accordance with the quality information. The compressed representation may be decoded without using the quality information.

Claims

1. A method for performing lossy encoding comprising:

receiving source data and quality information; and

mapping the source data into a compressed representation having a level of distortion in accordance with the quality information, wherein said compressed representation may be decoded without using said quality information.

2. The method of claim 1, wherein the compressed representation is formed using a fixed codebook with a variable partition.

3. The method of claim 1, wherein the quality information includes a distortion weighting value for each source sample included in said source data, said weighting value weighting an error between said each source sample and a lossy reconstruction of said source sample.

4. The method of claim 1, wherein the compressed representation is formed using a lattice.

5. The method of claim 1, wherein the compressed representation is formed using nested lattices.

6. The method of claim 1, wherein the compressed representation is formed using a transform.

7. The method of claim 1, wherein the compressed representation is formed using a curve fitting technique.

8. The method of claim 1, wherein the compressed representation is formed using an error correcting code.

9. The method of claim 1, wherein the compressed representation is formed using a hierarchy of one or more lattices and a linear code.

10. The method of claim 9, wherein a first source sample included in said source data is indicated by said quality information as more important than a second source sample included in said source data, and said first source sample is assigned a first lattice in said hierarchy and said second source sample is assigned a second different lattice in said hierarchy.

11. The method of claim 10, further comprising:

determining a quantization point in a lattice assigned to a source sample included in said source data; and

decomposing said quantization point for said source sample in said source data in accordance with quality information for said source sample.

12. The method of claim 11, wherein said decomposing further comprises:

performing a number of iterations of a decomposition step in accordance with a specified level of quality information for said source sample.

13. The method of claim 10, further comprising:

quantizing all source samples of said source data of a certain quality level as indicated by the quality data in accordance with an assigned lattice of quantization points;

decomposing each source sample of said certain quality level into one or more decomposition components; and

performing a first technique to correctly match a first decomposition component of each source sample of said certain quality level.

14. The method of claim 13, wherein said first technique interpolates a decomposition component for another source sample at a quality level other than said certain quality level.

15. The method of claim 1, wherein said quality information indicates an importance of each source sample included in said source data relative to other source samples included in said source data.

16. A method for performing lossy encoding comprising:

receiving a set of points and a set of quality information; and

mapping said set of points to quantized points, wherein a first rule is used for mapping a first of said points to a first quantized point in accordance with a fixed codebook and a first indicator in said quality information, a second rule is used for mapping a second of said points to a second quantized point in accordance with said codebook and a second indicator in said quality information.

17. The method of claim 16, wherein said quality information indicates an importance of each point in said set of points relative to other points included in said set of points.

18. The method of claim 17, wherein a rule is associated with each level of quality indicated in said quality information, said rule being used to determine a partition associated with each quantization point in said fixed codebook, said partition varying in accordance with said each level of quality.

19. The method of claim 18, wherein a first rule defines a quality level indicating that said first point is more important than said second point, said first rule defining partitions having a finer granularity than partitions defined by said second rule.

20. A method for performing lossy encoding comprising:

receiving a fixed codebook of quantization points and a plurality of rules, each rule defining a different partition used in mapping a source point to a quantization point;

receiving source data including a plurality of source points;

receiving quality information indicating a plurality of quality levels, each of said quality levels being associated with a different one of said rules, each of said source points being associated with one of said plurality of quality levels; and

determining a quantization point for each of said source points in accordance with a quality level and associated rule for each source point.

21. A computer program product for performing lossy encoding comprising code for:

receiving source data and quality information; and

mapping the source data into a compressed representation having a level of distortion in accordance with the quality information, wherein said compressed representation may be decoded without using said quality information.

22. The computer program product of claim 21, wherein the compressed representation is formed using a fixed codebook with a variable partition.

23. The computer program product of claim 21, wherein the quality information includes a distortion weighting value for each source sample included in said source data, said weighting value weighting an error between said each source sample and a lossy reconstruction of said source sample.

24. The computer program product of claim 21, wherein the compressed representation is formed using a lattice.

25. The computer program product of claim 21, wherein the compressed representation is formed using nested lattices.

26. The computer program product of claim 21, wherein the compressed representation is formed using a transform.

27. The computer program product of claim 21, wherein the compressed representation is formed using a curve fitting technique.

28. The computer program product of claim 21, wherein the compressed representation is formed using an error correcting code.

29. The computer program product of claim 21, wherein the compressed representation is formed using a hierarchy of one or more lattices and a linear code.

30. The computer program product of claim 29, wherein a first source sample included in said source data is indicated by said quality information as more important than a second source sample included in said source data, and said first source sample is assigned a first lattice in said hierarchy and said second source sample is assigned a second different lattice in said hierarchy.

31. The computer program product of claim 30, further comprising code for:

determining a quantization point in a lattice assigned to a source sample included in said source data; and

decomposing said quantization point for said source sample in said source data in accordance with quality information for said source sample.

32. The computer program product of claim 31, wherein said code for decomposing further comprises code for:

performing a number of iterations of a decomposition step in accordance with a specified level of quality information for said source sample.

33. The computer program product of claim 30, further comprising code for:

quantizing all source samples of said source data of a certain quality level as indicated by the quality data in accordance with an assigned lattice of quantization points;

decomposing each source sample of said certain quality level into one or more decomposition components; and

performing a first technique to correctly match a first decomposition component of each source sample of said certain quality level.

34. The computer program product of claim 33, wherein said first technique interpolates a decomposition component for another source sample at a quality level other than said certain quality level.

35. The computer program product of claim 21, wherein said quality information indicates an importance of each source sample included in said source data relative to other source samples included in said source data.

36. A computer program product for performing lossy encoding comprising code for:

receiving a set of points and a set of quality information; and

mapping said set of points to quantized points, wherein a first rule is used for mapping a first of said points to a first quantized point in accordance with a fixed codebook and a first indicator in said quality information, a second rule is used for mapping a second of said points to a second quantized point in accordance with said codebook and a second indicator in said quality information.

37. The computer program product of claim 36, wherein said quality information indicates an importance of each point in said set of points relative to other points included in said set of points.

38. The computer program product of claim 37, wherein a rule is associated with each level of quality indicated in said quality information, said rule being used to determine a partition associated with each quantization point in said fixed codebook, said partition varying in accordance with said each level of quality.

39. The computer program product of claim 38, wherein a first rule defines a quality level indicating that said first point is more important than said second point, said first rule defining partitions having a finer granularity than partitions defined by said second rule.

40. A computer program product for performing lossy encoding comprising code for:

receiving a fixed codebook of quantization points and a plurality of rules, each rule defining a different partition used in mapping a source point to a quantization point;

receiving source data including a plurality of source points,

receiving quality information indicating a plurality of quality levels, each of said quality levels being associated with a different one of said rules, each of said source points being associated with one of said plurality of quality levels; and

determining a quantization point for each of said source points in accordance with a quality level and associated rule for each source point.