Back to results
US 12,309,398 B2Granted
Chroma intra prediction in video coding and decoding
AssigneeBritish Broadcasting Corporation
InventorsMarc Gorriz Blanch, Marta Mrak, Saverio Blasi
Life of this patent
Estimating the term…
Abstract
An intra-prediction mode is provided for chroma data, in which a neural network implemented attention module directs the encoding of a block of chroma data with respect to luma data for the collocated luma block.
Claims16
A method of decoding video data, the method comprising:
- extracting reference samples from reconstructed luma samples and reconstructed chroma samples;
- and
- constructing at least one block of chroma prediction samples from the reference samples, wherein the constructing of chroma prediction samples depends on the spatial location of the reference samples, and wherein the constructing of chroma prediction samples depends on the usage of an attention module, where the attention module is configured as a deep neural network, wherein the attention module is configured to calculate an attention matrix comprising weights representing a contribution to the prediction of a given output location, each weight corresponding to a reference sample of the extracted reference samples.
A decoder for decoding video data, comprising:
- a reference sample extractor for extracting reference samples from reconstructed luma samples and reconstructed chroma samples;
- and
- a chroma prediction samples constructor for constructing at least one block of chroma prediction samples from the reference samples, wherein the chroma prediction samples constructor is operable to construct chroma prediction samples depending on the spatial location of the reference samples, wherein the chroma prediction samples constructor is operable to construct chroma prediction samples depending on the usage of an attention module, wherein the attention module is configured as a deep neural network, wherein the attention module is configured to calculate an attention matrix comprising weights representing a contribution to the prediction of a given output location, each weight corresponding to a reference sample of the extracted reference samples.
Dependencies are inferred from the claim text.