Determining scalar quantizers for a signal based on a target distortion

• US 7,639,886 B1
• Filed: 10/04/2004
• Issued: 12/29/2009
• Est. Priority Date: 10/04/2004
• Status: Active Grant

• Current Assignee: Adobe Inc.
• Original Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Inventors: Rastogi, Anubha
• Primary Examiner(s): Bali; Vikkram
• Assistant Examiner(s): YEH, EUENG NAN
• Application Number:
  US10/958,802
• Time in Patent Office: 1,912 Days
• Field of Search: 382/251
• US Class Current: 382/251
• CPC Class Codes: H04N 19/115 Selection of the code volume for a coding unit prior to coding
• CPC Class Codes: H04N 19/124 Quantisation
• CPC Class Codes: H04N 19/146 Data rate or code amount at the encoder output
• CPC Class Codes: H04N 19/154 Measured or subjectively estimated visual quality after decoding, e.g. measurement of distortion use of rate-distortion criteria H04N19/147
• CPC Class Codes: H04N 19/1883 the unit relating to sub-band structure, e.g. hierarchical level, directional tree, e.g. low-high [LH], high-low [HL], high-high [HH]
• CPC Class Codes: H04N 19/34 Scalability techniques involving progressive bit-plane based encoding of the enhancement layer, e.g. fine granular scalability [FGS]
• CPC Class Codes: H04N 19/63 using sub-band based transform, e.g. wavelets
• CPC Class Codes: H04N 19/647 using significance based coding, e.g. Embedded Zerotrees of Wavelets [EZW] or Set Partitioning in Hierarchical Trees [SPIHT]
• 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 19/85 using pre-processing or post-processing specially adapted for video compression

Abstract

Methods and apparatus implementing systems and techniques for determining quantizer values for a signal. In general, in one implementation, a technique includes: obtaining a target distortion and a transformed signal having sub-bands resulting from a source signal being transformed by a wavelet-based decomposition; distributing the target distortion among the sub-bands based on energy weights to produce target sub-band distortions; ascertaining actual sub-band distortion values that result at different grades of quantization; determining quantizer values for the sub-bands, including interpolating between at least two of the quantization grades; and quantizing the sub-bands with the determined quantizer values. The signal may be a digital image. Additionally, in another implementation, determining the quantizer values can involve determining approximations of sub-band coefficient quantization error rather than performing the interpolation, in which case, the ascertaining operation need not be performed.

Claims

1. A method comprising:

obtaining a target distortion and a transformed signal, the target distortion indicating desired signal quality, and the transformed signal comprising sub-components resulting from a source signal being transformed by a multi-resolution decomposition;

distributing the target distortion among the sub-components to produce target sub-component distortions;

determining quantizer values for the sub-components based on the target sub-component distortions; and

quantizing the sub-components with the determined quantizer values;

wherein the distributing, the determining and the quantizing are performed using one or more processors;

wherein the sub-components comprise sub-bands resulting from the source signal being transformed by a wavelet-based, frequency-dependent, multi-resolution decomposition, the target sub-component distortions comprise target sub-band distortions, and distributing the target distortion comprises distributing the target distortion among the sub-bands based on energy weights of the sub-bands;

the method further comprising ascertaining actual sub-band distortion values that result at different grades of quantization in a current sub-band, wherein determining quantizer values comprises;

if a target sub-band distortion for the current sub-band falls between at least two of the actual sub-band distortion values determined for the current sub-band, interpolating between at least two of the quantization grades based on the at least two actual sub-band distortion values to determine a quantizer value for the current sub-band corresponding to the target sub-band distortion for the current sub-band; and

wherein distributing the target distortion comprises distributing the target distortion among the sub-bands in inverse proportion to the energy weights of the sub-bands, and determining quantizer values further comprises;

determining quantizer values for the sub-bands starting at a highest resolution level and ending at a lowest resolution level of the sub-bands, and

if one or more sub-bands of a current resolution level cannot satisfy one or more corresponding target sub-band distortions, redistributing an unsatisfied portion of the one or more corresponding target sub-band distortions to target sub-band distortions for lower resolution sub-bands in inverse proportion to energy weights of the lower resolution sub-bands.

2. The method of claim 1, further comprising removing one or more least-significant bit levels in binary sub-band data to produce at least one of the different grades of quantization corresponding to exponential-powers-of-two quantizers.

3. The method of claim 1, further comprising:

presenting a preview of the quantized, transformed signal to demonstrate a result of the target distortion on the transformed signal; and

repeating the distributing, determining, quantizing and presenting operations, but not the ascertaining operation, in response to a change in the target distortion.

4. The method of claim 1, further comprising:

compressing the quantized, transformed signal to form a compressed signal; and

applying a post-compression rate-distortion optimization on the compressed signal.

5. A software product tangibly embodied in a computer-readable medium, the software product comprising instructions operable to cause data processing apparatus to perform operations comprising:

obtaining a target distortion and a transformed signal, the target distortion indicating desired signal quality, and the transformed signal comprising sub-components resulting from a source signal being transformed by a multi-resolution decomposition;

distributing the target distortion among the sub-components to produce target sub-component distortions;

determining quantizer values for the sub-components based on the target sub-component distortions; and

quantizing the sub-components with the determined quantizer values;

wherein the sub-components comprise sub-bands resulting from the source signal being transformed by a wavelet-based, frequency-dependent, multi-resolution decomposition, the target sub-component distortions comprise target sub-band distortions, and distributing the target distortion comprises distributing the target distortion among the sub-bands based on energy weights of the sub-bands;

wherein the operations further comprise ascertaining actual sub-band distortion values that result at different grades of quantization in a current sub-band, and wherein determining quantizer values comprises;

if a target sub-band distortion for the current sub-band falls between at least two of the actual sub-band distortion values determined for the current sub-band, interpolating between at least two of the quantization grades based on the at least two actual sub-band distortion values to determine a quantizer value for the current sub-band corresponding to the target sub-band distortion for the current sub-band; and

wherein distributing the target distortion comprises distributing the target distortion among the sub-bands in inverse proportion to the energy weights of the sub-bands, and determining quantizer values further comprises;

determining quantizer values for the sub-bands starting at a highest resolution level and ending at a lowest resolution level of the sub-bands, and

if one or more sub-bands of a current resolution level cannot satisfy one or more corresponding target sub-band distortions, redistributing an unsatisfied portion of the one or more corresponding target sub-band distortions to target sub-band distortions for lower resolution sub-bands in inverse proportion to energy weights of the lower resolution sub-bands.

6. The software product of claim 5, wherein the operations further comprise removing one or more least-significant bit levels in binary sub-band data to produce at least one of the different grades of quantization corresponding to exponential-powers-of-two quantizers.

7. The software product of claim 6, wherein the transformed signal comprises a two-dimensional data set, and removing one or more least-significant bit levels comprises removing one or more least-significant bit planes.

8. The software product of claim 5, wherein the operations further comprise:

presenting a preview of the quantized, transformed signal to demonstrate a result of the target distortion on the transformed signal; and

repeating the distributing, determining, quantizing and presenting operations, but not the ascertaining operation, in response to a change in the target distortion.

9. The software product of claim 5, wherein interpolating comprises employing quadratic interpolation.

10. The software product of claim 5, wherein the operations further comprise:

compressing the quantized, transformed signal to form a compressed signal; and

applying a post-compression rate-distortion optimization on the compressed signal.

11. The software product of claim 5, wherein obtaining the target distortion comprises generating a target peak-signal-to-noise ratio from a received quality measure.

12. The software product of claim 5, wherein obtaining the transformed signal comprises receiving at least a portion of a digital image.

13. A method comprising:

obtaining a target distortion and a signal, the target distortion indicating desired signal quality;

ascertaining actual distortion values that result at different grades of quantization of the signal;

interpolating between at least two of the quantization grades based on at least two of the actual distortion values to determine a quantizer value corresponding to the target distortion;

quantizing the signal with the determined quantizer value;

presenting a preview of the quantized signal to demonstrate a result of the target distortion on the signal; and

repeating the interpolating, quantizing and presenting operations, but not the ascertaining operation, in response to a change in the target distortion;

wherein the interpolating, the quantizing, the presenting and the repeating are performed using one or more processors.

14. The method of claim 13, further comprising removing one or more least-significant bit levels in binary signal data to produce at least one of the different grades of quantization corresponding to exponential-powers-of-two quantizers.

15. The method of claim 13, wherein the signal comprises a first sub-band of a larger transformed signal comprising multiple sub-bands, including the first sub-band, resulting from a source signal being transformed by a wavelet-based decomposition, and obtaining the target distortion for the first sub-band comprises:

receiving a larger target distortion for the larger transformed signal; and

distributing the larger target distortion among the multiple sub-bands to produce target sub-band distortions, including the target distortion for the first sub-band.

16. The method of claim 15, wherein distributing the larger target distortion comprises distributing the larger target distortion among the multiple sub-bands based on energy weights of the multiple sub-bands.

17. The method of claim 16, further comprising redistributing an unsatisfied portion of a target sub-band distortion for a current sub-band to subsequent sub-bands based on energy weights of the subsequent sub-bands.

18. The method of claim 13, further comprising:

compressing the quantized, transformed signal to form a compressed signal; and

applying a post-compression rate-distortion optimization on the compressed signal.

19. A software product tangibly embodied in a computer-readable medium, the software product comprising instructions operable to cause data processing apparatus to perform operations comprising:

obtaining a target distortion and a signal, the target distortion indicating desired signal quality;

ascertaining actual distortion values that result at different grades of quantization of the signal;

interpolating between at least two of the quantization grades based on at least two of the actual distortion values to determine a quantizer value corresponding to the target distortion;

quantizing the signal with the determined quantizer value;

presenting a preview of the quantized signal to demonstrate a result of the target distortion on the signal; and

repeating the interpolating, quantizing and presenting operations, but not the ascertaining operation, in response to a change in the target distortion.

20. The software product of claim 19, wherein the operations further comprise removing one or more least-significant bit levels in binary signal data to produce at least one of the different grades of quantization corresponding to exponential-powers-of-two quantizers.

21. The software product of claim 20, wherein the signal comprises a two-dimensional data set, and removing one or more least-significant bit levels comprises removing one or more least-significant bit planes.

22. The software product of claim 19, wherein the signal comprises a first sub-band of a larger transformed signal comprising multiple sub-bands, including the first sub-band, resulting from a source signal being transformed by a wavelet-based decomposition, and obtaining the target distortion for the first sub-band comprises:

receiving a larger target distortion for the larger transformed signal; and

distributing the larger target distortion among the multiple sub-bands to produce target sub-band distortions, including the target distortion for the first sub-band.

23. The software product of claim 22, wherein distributing the larger target distortion comprises distributing the larger target distortion among the multiple sub-bands based on energy weights of the multiple sub-bands.

24. The software product of claim 23, wherein the operations further comprise redistributing an unsatisfied portion of a target sub-band distortion for a current sub-band to subsequent sub-bands based on energy weights of the subsequent sub-bands.

25. The software product of claim 19, wherein the operations further comprise:

compressing the quantized, transformed signal to form a compressed signal; and

applying a post-compression rate-distortion optimization on the compressed signal.

26. The software product of claim 19, wherein interpolating comprises employing quadratic interpolation.

27. The software product of claim 19, wherein obtaining the target distortion comprises generating a target peak-signal-to-noise ratio from a received quality measure.

28. The software product of claim 19, wherein obtaining the signal comprises receiving at least a portion of a digital image.

29. A system comprising:

a data processing apparatus; and

a machine-readable medium tangibly embodying a software product comprising instructions operable to cause the data processing apparatus to present an imaging software user interface and to perform operations comprising;

obtaining a target distortion and a transformed image, the target distortion indicating desired image quality, and the transformed image comprising sub-bands resulting from a source image being transformed by a wavelet-based decomposition;

distributing the target distortion among the sub-bands based on energy weights of the sub-bands to produce target sub-band distortions;

ascertaining, for at least four of the sub-bands, actual sub-band distortion values that result at different grades of quantization;

determining quantizer values for the sub-bands based on the target sub-band distortions, including, for each of the at least four sub-bands, interpolating between at least two of the quantization grades based on at least two corresponding actual sub-band distortion values to determine a quantizer value for a current sub-band corresponding to a target sub-band distortion for the current sub-band; and

quantizing the sub-bands with the determined quantizer values;

wherein distributing the target distortion comprises redistributing an unsatisfied portion of a target sub-band distortion for a current sub-band to subsequent sub-bands based on energy weights of the subsequent sub-bands.

30. The system of claim 29, wherein the operations further comprise:

presenting a preview to demonstrate a result of the quantization according to the target distortion;

repeating the interpolating, quantizing and presenting operations, but not the ascertaining operation, in response to a change in the target distortion; and

compressing the quantized sub-bands.