Methods and apparatus for video encoding and decoding using adaptive geometric partitioning

Patent metadata

Field	Value
Publication number	US-9326003-B2
Application number	US-201013379863-A
Country	US
Kind code	B2
Filing date	Jun 25, 2010
Priority date	Jun 26, 2009
Publication date	Apr 26, 2016
Grant date	Apr 26, 2016

Abstract

Methods and apparatus are provided for video encoding and decoding using adaptive geometric partitioning. An apparatus includes an encoder for encoding picture data for a region in a picture using adaptive geometric partitioning, wherein a subset of geometric partitions used to code the picture data for the region is determined responsive to at least one of a distance and an angle. The distance is measured from a defined point within the region to a partition boundary. The angle is measured with respect to at least a line extending from the defined point within the region to the partition boundary.

First claim

The invention claimed is:

1. An apparatus, comprising:
a video encoder, comprising a processor and a memory connected thereto, configured to encode picture data for a region in a picture using adaptive geometric partitioning, wherein an initial subset of geometric partitions used to code the picture data for the region is determined responsive to at least one of a distance and an angle, the distance measured from a point within the region to a partition boundary, and the angle measured with respect to a line extending from the point within the region to the partition boundary, and wherein an initial subset of geometric partitions to use to code picture data for a region is determined by analyzing statistics of partitions previously selected and coded, refining the subset, and coding additional regions, wherein numbers of regions to analyze and code are specified in at least one high level syntax element, the region being encoded by evaluating only geometric partitions in the subset, the subset including less than all available geometric partitions, and wherein the subset of geometric partitions is based on using non-uniform sampling of the distance and a density of the angle that is a function of the distance from a defined point of a region to a partition boundary and by limiting the distance and angle sampling space,
and wherein angles are denser when a distance from the defined point to a center of the block is small and angles are sparser when that distance from the defined point to a center of the block is larger.

2. In a video encoder, comprising a processor and a memory connected thereto, configured to encode, a method, comprising:
encoding picture data for a region in a picture using adaptive geometric partitioning, wherein an initial subset of geometric partitions used to code the picture data for the region is determined responsive to at least one of a distance and an angle, the distance measured from a point within the region to a partition boundary, and the angle measured with respect to a line extending from the point within the region to the partition boundary, and wherein an initial subset of geometric partitions to use to code picture data for a region is determined by analyzing statistics of partitions previously selected and coded, refining the subset, and coding additional regions, wherein numbers of regions to analyze and code are specified in at least one high level syntax element, the region being encoded by evaluating only geometric partitions in the subset, the subset including less than all available geometric partitions, and wherein the subset of geometric partitions is based on using non-uniform sampling of the distance and a density of the angle that is a function of the distance from a defined point of a region to a partition boundary and by limiting the distance and angle sampling space,
and wherein angles are denser when a distance from the defined point to a center of the block is small and angles are sparser when that distance from the defined point to a center of the block is larger.

3. The method of claim 2 , wherein the point within the region is the center of the region, and the subset of geometric partitions is determined responsive to the distance from the center of the region to the partition boundary.

4. The method of claim 3 , wherein a distance step and an angle step are used to sample a distance space and an angle space for the adaptive geometric partitioning to construct a geometric partition dictionary, and the distance step and the angle step are a function of the distance from the point to the partition boundary, the subset of geometric partitions having less geometric partitions than the geometric partition dictionary.

5. The method of claim 3 , wherein the subset of geometric partitions includes geometric partitions having the distance from the point to the partition boundary equal to or smaller than a threshold distance.

6. The method of claim 2 , wherein the subset of geometric partitions is based on previously coded information of at least one of the region and previously coded other regions in the picture.

7. The method of claim 6 , wherein the subset of geometric partitions is based on a partition selected for the previously coded other regions.

8. The method of claim 7 , wherein a partition selected for the region is extrapolated to predict a partition for a subsequent region, and a subset of geometric partitions for coding the subsequent region is based on the predicted partition for the subsequent region.

9. The method of claim 6 , wherein the subset of geometric partitions is derived based on statistics of partitions selected for the previously coded other regions.

10. The method of claim 9 , wherein the picture is comprised in a video sequence, a set of pictures in the video sequence, each excluding the region, are encoded using a default dictionary of geometric partitions corresponding to a brute force partition computation approach, and statistics of partitions selected for the set of pictures are used to construct the subset of geometric partitions for the region, the subset having less geometric partitions than the default dictionary.

11. The method of claim 2 , wherein the subset of geometric partitions made available for the adaptive geometric partitioning is user defined and explicitly signaled in a resulting bitstream.

12. The method of claim 2 , wherein the subset is one of a plurality of subsets, wherein the subset and the plurality all together represent less than all possible partitions, the plurality of subsets being pre-stored, and wherein the method further comprises sending an index to a decoder using one or more high level syntax elements, the index indicating a particular one of more of the plurality of subsets to be used for decoding.

13. The method of claim 2 , further comprising transmitting, in a bitstream to a decoder, the subset of geometric partitions for use by the decoder in decoding at least the picture data for the region.

14. The method of claim 2 , wherein said encoding step comprises:
selecting a partition for the region from a default dictionary of geometric partitions based on rate-distortion criteria; and
performing geometric motion compensation to obtain a residue for the region, and
reconstructing the residue such that only the selected partition, from among multiple available partitions in the default dictionary of geometric partitions, is capable of being inferred from a subsequent reconstruction of the residue at a decoder.

15. An apparatus, comprising:
a video decoder, comprising a processor and memory connected thereto, configured to decode picture data for a region in a picture using adaptive geometric partitioning, wherein an initial subset of geometric partitions used to decode the picture data for the region is determined responsive to at least one of a distance and an angle, the distance measured from a point within the region to a partition boundary, and the angle measured with respect to a line extending from the point within the region to the partition boundary, and wherein an initial subset of geometric partitions to use to decode picture data for a region is determined by analyzing statistics of partitions previously selected and decoded, refining the subset, and decoding additional regions, wherein numbers of regions to analyze and code are specified in at least one high level syntax element, the region being decoded by evaluating only geometric partitions in the subset, the subset including less than all available geometric partitions, and wherein the subset of geometric partitions is based on using non-uniform samplin

Assignees

• Guo Liwei
• Yin Peng
• Sole Joel
• Zheng Yunfei
• Lu Xiaoan
• Xu Qian
• Thomson Licensing

Inventors

• Guo Liwei
• Yin Peng
• Sole Joel
• Zheng Yunfei
• Lu Xiaoan
• Xu Qian

Classifications

• H04N19/176

  the region being a block, e.g. a macroblock · CPC title

• H04N19/11

  among a plurality of spatial predictive coding modes · CPC title

• H04N19/147

  according to rate distortion criteria (rate-distortion as a criterion for motion estimation H04N19/567) · CPC title

• H04N19/146

  Data rate or code amount at the encoder output · CPC title

• H04N19/109

  among a plurality of temporal predictive coding modes · CPC title