Multi-level significance maps for encoding and decoding

What is this patent about?

’782 is related to the field of video data compression, specifically the efficient entropy encoding and decoding of quantized transform coefficients. In modern video standards like HEVC, significance maps are used to identify the locations of non-zero coefficients within a transform unit, but as transform sizes increase to 16x16 or 32x32, the computational cost of managing individual coefficient contexts and the bitstream overhead for sparse data become significant bottlenecks.

The underlying idea behind ’782 is the implementation of a multi-level significance map structure that partitions a transform unit into smaller, non-overlapping blocks or groups. By first signaling a high-level flag for each group to indicate whether it contains any non-zero data, the system can bypass the encoding of individual flags for entirely empty regions. This hierarchical approach reduces memory bandwidth requirements and improves compression efficiency by avoiding the processing of large clusters of zero-value coefficients.

The claims of ’782 focus on a specific encoding logic that utilizes these significant-coefficient-group flags to control the bitstream inclusion of individual coefficient flags. A critical aspect of the claimed mechanism is an inference rule for the flag at position (0,0) within a group: if a group is signaled as non-empty but all previously scanned flags in that group are zero, the final flag is inferred to be one. This eliminates the need to explicitly encode the last possible non-zero flag in a group, further tightening the bitstream.

In practice, the invention operates by scanning the transform unit in a prescribed order, such as a reverse diagonal scan, to determine which groups contain significant data. The encoder uses context-adaptive binary arithmetic coding (CABAC) with contexts derived from the status of neighboring groups, such as those to the right or below the current block. This spatial dependency allows the probability model to adapt to the typical clustering of energy in the low-frequency regions of a transform block.

This approach differentiates itself from prior methods by replacing computationally intensive neighbor-based context derivation for every single coefficient with a more streamlined hierarchical signaling strategy. By incorporating rate-distortion optimized quantization (RDOQ) at the group level, the encoder can strategically zero out entire blocks if the bit savings outweigh the visual distortion. This results in a high-performance coding engine that scales effectively with the larger transform units required for high-definition video content.