Methods and apparatus for video encoding and decoding using motion matrix

Patent metadata

Field	Value
Publication number	US-9894354-B2
Application number	US-201113879349-A
Country	US
Kind code	B2
Filing date	Oct 10, 2011
Priority date	Oct 14, 2010
Publication date	Feb 13, 2018
Grant date	Feb 13, 2018

Abstract

Methods and apparatus are provided for video encoding and decoding using a motion matrix. An apparatus includes a video encoder for encoding a picture in a video sequence using a motion matrix. The motion matrix has a rank below a given threshold and a sparse representation with respect to a dictionary. The dictionary includes a set of atoms and basis vectors for representing the picture and for permitting the picture to be derived at a corresponding decoder using only the set. The dictionary formed from a set of reference pictures in the video sequence.

First claim

The invention claimed is:

1. An apparatus, comprising:
a video encoder for encoding a picture in a video sequence using a motion matrix comprising data enabling a decoder to uniquely reconstruct the picture using said motion matrix and reference frames, wherein the motion matrix has a rank below a given value and a sparse representation with respect to a dictionary, said dictionary selected from among a plurality of dictionaries generated as a function of at least one reference frame and not of said picture being encoded, and said selection is based on achieving a target rank and target sparsity and said motion matrix is found by iteratively finding a representation of said picture in terms of said selected dictionary, keeping only said target sparsity number of largest entries of said representation, updating the motion matrix and keeping the largest target rank number of Eigenvalues of the motion matrix up to said target rank number until the representation converges to a particular level.

2. The apparatus of claim 1 , wherein the dictionary is formed from spatial and temporal correlations between the picture and the set of reference pictures.

3. The apparatus of claim 1 , wherein the video encoder comprises an embedding device for embedding the motion matrix into an embedded matrix, a sampler for sampling the embedded matrix to provide a subset of samples from the embedded matrix, a quantizer for quantizing the subset of samples to provide a quantized subset of samples, and an entropy coder for entropy coding the quantized subset of samples into compressed symbols.

4. The apparatus of claim 3 , wherein the embedded matrix is determined using two matrices which are multiplied by the motion matrix, the two matrices being selected to provide a particular amount of random Eigen-vectors in the embedded matrix.

5. The apparatus of claim 3 , wherein the subset of samples are randomly selected from the embedded matrix.

6. The apparatus of claim 5 , wherein a same random seed and a same random number generator are used by the video encoder and a corresponding video decoder to ensure that the samples in the subset of samples from the embedded matrix have same respective locations at the video encoder and at the corresponding video decoder.

7. The apparatus of claim 3 , wherein said sampler discards portions of the embedded matrix having different locations in the embedded matrix than the samples in the subset of samples.

8. In a video encoder, a method, comprising:
encoding a picture in a video sequence using a motion matrix, comprising data enabling a decoder to uniquely reconstruct the picture using said motion matrix and reference frames, wherein the motion matrix has a rank below a given value and a sparse representation with respect to a dictionary, said dictionary selected from among a plurality of dictionaries generated as a function of at least one reference frame and not of said picture being encoded, and said selection is based on achieving a target rank and target sparsity and said motion matrix is found by iteratively finding a representation of said picture in terms of said selected dictionary, keeping only said target sparsity number of largest entries of said representation, updating the motion matrix and keeping the largest target rank number of Eigenvalues of the motion matrix up to said target rank number until the representation converges to a particular level.

9. The method of claim 8 , wherein the dictionary is formed from spatial and temporal correlations between the picture and the set of reference pictures.

10. The method of claim 8 , wherein said encoding step comprises:
embedding the motion matrix into an embedded matrix;
sampling the embedded matrix to provide a subset of samples from the embedded matrix;
quantizing the subset of samples to provide a quantized subset of samples; and
entropy coding the quantized subset of samples into compressed symbols.

11. The method of claim 10 , wherein the embedded matrix is determined using two matrices which are multiplied by the motion matrix, the two matrices being selected to provide a particular amount of random Eigen-vectors in the embedded matrix.

12. The method of claim 10 , wherein the subset of samples are randomly selected from the embedded matrix.

13. The method of claim 12 , wherein a same random seed and a same random number generator are used by the video encoder and a corresponding video decoder to ensure that the samples in the subset of samples from the embedded matrix have same respective locations at the video encoder and at the corresponding video decoder.

14. The method of claim 10 , wherein said sampling step comprises discarding portions of the embedded matrix having different locations in the embedded matrix than the samples in the subset of samples.

15. An apparatus, comprising:
a video decoder for decoding a picture in a video sequence using an approximation of a motion matrix, comprising data enabling a decoder to uniquely reconstruct the picture using said motion matrix and reference frames wherein the approximation of the motion matrix has a rank below a given value and a sparse representation with respect to a dictionary, said dictionary selected from among a plurality of dictionaries generated as a function of at least one reference frame and not of said picture being encoded, and said selection is based on achieving a target rank and target sparsity and said motion matrix is found by iteratively finding a representation of said picture in terms of said selected dictionary, keeping only said target sparsity number of largest entries of said representation, updating the motion matrix and keeping the largest target rank number of Eigenvalues of the motion matrix up to said target rank number until the representation converges to a particular level.

16. The apparatus of claim 15 , wherein the dictionary is formed from spatial and temporal correlations between the picture and the set of reference pictures.

17. The apparatus of claim 15 , wherein the video decoder comprises an entropy decoder for entropy decoding compressed symbols to provide a quantized subset of samples from an embedded matrix, a de-quantizer for de-quantizing the quantized subset of samples to provide an approximation of a subset of samples from a recovered version of the embedded matrix, a matrix completion device for determining missing portions of the recovered version of the embedded matrix to provide a completed recovered version of the embedded matrix, and a reconstruction device for reconstructing the approximation of the motion matrix from the completed recovered version of the embedded matrix.

18. The apparatus of claim 17 , wherein the missing portions of the recovered version of the embedded matrix correspond to previously discarded portions of the embedded matrix having different locations in the embedded matrix than the samples in the subset of samples.

19. The apparatus of claim 17 , wherein the approximation of the motion matrix is determined using the embedded matrix and two matrices which are multiplied by the embedded matrix matrix.

20. The apparatus of claim 17 , wherein a same random seed and a same random number generator are used by the video decoder and a corresponding video encoder to ensure that the samples in the subset of samples from the embedded matrix have same respective locations at the video decoder and at the corresponding video encoder.

21. The apparatus of claim 15 , wherein a sparse representation of the motion matri

Assignees

• Moghadam Abdolreza
• Sole Joel
• Lu Xiaoan
• Thomson Licensing

Inventors

• Moghadam Abdolreza
• Sole Joel
• Lu Xiaoan

Classifications

• H04N19/517

  by encoding · CPC title

• H04N19/97 Primary

  Matching pursuit coding · CPC title

• H04N19/172

  the region being a picture, frame or field · CPC title

• H04N19/132

  Sampling, masking or truncation of coding units, e.g. adaptive resampling, frame skipping, frame interpolation or high-frequency transform coefficient masking · CPC title

• H04N19/0069 Primary

  Electricity · mapped topic