Selectively using multiple entropy models in adaptive coding and decoding

• US 20070016418A1
• Filed: 07/15/2005
• Published: 01/18/2007
• Est. Priority Date: 07/15/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Chen, Wei-Ge, Mehrotra, Sanjeev
• Granted Patent: US 7,599,840 B2
• Time in Patent Office: Days
• US Class Current: 704/240
• CPC Class Codes: G10L 19/0017 Lossless audio signal coding; Perfect reconstruction of coded audio signal by transmission of coding error G10L19/24 takes precedence
• CPC Class Codes: H03M 7/40 Conversion to or from variable length codes, e.g. Shannon-Fano code, Huffman code, Morse code
• CPC Class Codes: H04N 19/13 Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC]
• CPC Class Codes: H04N 19/46 Embedding additional information in the video signal during the compression process H04N19/517, H04N19/68, H04N19/70 take precedence

Abstract

Techniques and tools for selectively using multiple entropy models in adaptive coding and decoding are described herein. For example, for multiple symbols, an audio encoder selects an entropy model from a first model set that includes multiple entropy models. Each of the multiple entropy models includes a model switch point for switching to a second model set that includes one or more entropy models. The encoder processes the multiple symbols using the selected entropy model and outputs results. Techniques and tools for generating entropy models are also described.

Claims

1. A method comprising:

for plural symbols, selecting an entropy model from a first model set that includes multiple entropy models, each of the multiple entropy models of the first model set including a model switch point for switching to a second model set that includes one or more entropy models;

processing the plural symbols using the selected entropy model; and

outputting results of the processing.

2. The method of claim 1 wherein an encoder performs the selecting, the processing and the outputting during encoding, wherein the processing includes entropy encoding.

3. The method of claim 1 wherein a decoder performs the selecting, the processing and the outputting during decoding, wherein the processing includes entropy decoding.

4. The method of claim 1 wherein the multiple entropy models of the first model set and the one or more entropy models of the second model set are probability distributions for arithmetic coding and/or decoding, and wherein the model switch point is a model switch probability in the multiple probability distributions of the first model set.

5. The method of claim 1 wherein the multiple entropy models of the first model set are embodied respectively in multiple VLC tables of a first table set, wherein the one or more entropy models of the second model set are embodied respectively in one or more VLC tables of a second table set, wherein the model switch point is an escape code, and wherein each of the multiple VLC tables of the first table set includes the escape code for switching to the second table set.

6. The method of claim 5 wherein the multiple VLC tables of the first table set and the one or more VLC tables of the second table set are Huffman code tables, and wherein the second table set includes a single Huffman code table, such that the single Huffman code table represents a common branch in trees representing the respective multiple Huffman code tables of the first table set.

7. The method of claim 5 wherein the multiple VLC tables of the first table set are adapted for a first symbol value set that includes more probable symbol values, and wherein the one or more VLC tables of the second table set are adapted for a second symbol value set that includes less probable symbol values.

8. The method of claim 7 wherein the second table set includes a single VLC table, and wherein the processing is for two-stage variable length coding or decoding of those of the plural symbols having the less probable symbol values.

9. The method of claim 1 further comprising generating the multiple entropy models of the first model set and the one or more entropy models of the second model set, wherein the generating includes:

clustering probability distributions according to a first cost metric, resulting in plural preliminary clusters; and

refining the plural preliminary clusters according to a second cost metric different than the first cost metric, resulting in plural final clusters.

10. The method of claim 1 wherein the second model set includes a single entropy model, the method further comprising generating the multiple entropy models of the first model set and the single entropy model of the second model set, wherein the generating includes, for the single entropy model of the second model set, constraining less probable symbol values to have a common conditional distribution across probability distributions.

11. The method of claim 1 wherein each of the one or more entropy models of the second model set includes a second model switch point for switching to a third model set that includes one or more entropy models.

12. The method of claim 1 wherein, for at least some of the multiple entropy models of the first model set, the model switch point has a different value from model to model.

13. The method of claim 1 wherein each of the multiple entropy models of the first model set further includes a second model switch point for switching to a third model set that includes one or more entropy models.

14. The method of claim 1 wherein the plural symbols are for quantized spectral coefficients for audio data.

15. The method of claim 1 wherein the selecting is part of forward adaptive switching.

16. The method of claim 1 wherein the selecting is part of backward adaptive switching.

17. A system comprising one or more modules for generating entropy models by:

clustering probability distributions according to a first cost metric, resulting in plural preliminary clusters;

refining the plural preliminary clusters according to a second cost metric different than the first cost metric, resulting in plural final clusters; and

setting the entropy models based at least in part upon the plural final clusters.

18. The system of claim 17 wherein the second cost metric is relative entropy.

19. A system comprising:

means for obtaining probability distributions for symbol values; and

means for generating entropy models, including constraining plural less probable symbol values to have a common conditional distribution across the probability distributions without so constraining plural more probable symbol values.

20. The system of claim 19 wherein the entropy models are respectively embodied in multiple VLC tables of a first table set and a single VLC table of a second table set, wherein the multiple VLC tables are adapted for the plural more probable symbol values, and wherein the single VLC table is adapted for the plural less probable symbol values.