Multi-Level Significance Maps For Encoding And Decoding

What is this patent about?

’758 is related to the field of video encoding and decoding, specifically addressing the efficient representation of transform coefficient data. Modern video codecs, like H.264/AVC and HEVC, transform video frames into blocks of coefficients, quantize them, and then entropy encode the resulting data. A significant portion of the encoded data represents the locations of non-zero coefficients within these blocks, known as the significance map. Efficiently encoding this map is crucial for achieving high compression ratios.

The underlying idea behind ’758 is to introduce a multi-level significance map approach. Instead of directly encoding each individual coefficient's significance, the invention groups coefficients into blocks and uses a higher-level flag to indicate whether a block contains any non-zero coefficients. This allows the encoder to skip encoding the significance of individual coefficients within blocks that are entirely zero, thereby reducing the overall bit rate.

The claims of ’758 focus on a method, encoder, and non-transitory medium for determining whether to encode a value of a significant-coefficient flag corresponding to a transform coefficient of a block among one or more non-overlapping blocks within a transform unit. The determination is based on a context derived from the values of significant-coefficient-group flags of blocks adjacent to the right and below the current block. If the significant-coefficient-group flag is one, the significant-coefficient flag is encoded into the bitstream, otherwise, it is skipped.

In practice, the encoder first divides the transform unit into non-overlapping blocks. For each block, it determines a context for the significant-coefficient-group flag based on the flags of its right and bottom neighbors. If both neighbors have zero flags, a first context is used; otherwise, a second context is used. The significant-coefficient-group flag is then encoded using this context. If the group flag is one, the encoder proceeds to encode the individual coefficient flags within that block, unless all other flags in the group have been previously determined to be zero, in which case encoding is skipped. If the group flag is zero, all coefficient flags within that block are skipped.

This approach differs from prior methods by introducing a hierarchical representation of the significance map. Instead of directly encoding each coefficient's significance, ’758 uses a group-level flag to indicate the presence of non-zero coefficients within a block. This allows the encoder to efficiently skip encoding entire blocks of zero coefficients, leading to improved compression performance, especially for larger transform units where the probability of having all-zero blocks is higher. The context modeling based on neighboring blocks further enhances the coding efficiency by adapting to local statistics.