Significance map encoding and decoding using partition set based context assignment

What is this patent about?

’630 is related to the field of video data compression, specifically the entropy encoding and decoding of significance maps. In modern codecs like HEVC, significance maps indicate the locations of non-zero coefficients within a transform unit, and their efficient compression is vital as they represent a substantial portion of the total bitstream. Traditional methods often assign contexts to these maps in a uniform or rigid spatial manner, which can lead to suboptimal probability estimation and slow adaptation during the arithmetic coding process.

The underlying idea behind ’630 is that the statistical distribution of non-zero coefficients in a transform block is typically non-uniform, with energy concentrated in the low-frequency regions. By employing non-spatially-uniform partitioning, the invention allows the encoder and decoder to group bit positions into context parts that better reflect actual data usage. This approach balances the need for high-resolution modeling in high-probability areas with the need for faster context adaptation in areas where data is sparse, such as the high-frequency bottom-right regions of a block.

The claims of ’630 focus on a specific method and apparatus for reconstructing a significance map for a 4×4 transform unit using a predefined context mapping. The independent claims protect a precise block-based assignment where fifteen bit positions are mapped to nine distinct contexts (indexed 0 through 8). This specific geometric arrangement ensures that certain positions share a context—for instance, the mapping dictates that the third and seventh positions, as well as the eleventh, fourteenth, and fifteenth positions, are grouped together to optimize the probability state tracking.

In practice, the system determines the appropriate context for each bit position by looking up its coordinates in the defined partition set. As each bit is decoded via context-adaptive binary arithmetic coding, the associated probability state is updated immediately. This allows the decoder to refine its statistical model in real-time. The invention also contemplates switching between different partition sets—such as moving from a coarse to a fine partition—based on the size of the encoded slice or other sequence-level characteristics to maintain high efficiency across varying bitrates.

This approach differs from prior solutions by moving away from the one-context-per-position model or simple uniform quadrant-based grouping. By utilizing asymmetric partitions that cluster smaller parts in the top-left and larger parts toward the bottom-right, the invention reduces the total number of contexts the hardware must track while simultaneously improving compression gain. Furthermore, the invention suggests reordering the scan path to group identical contexts together, which minimizes context-switching overhead in high-speed hardware implementations.