Significance Map Encoding And Decoding Using Partition Selection

What is this patent about?

’210 is related to the field of video compression, specifically the encoding and decoding of significance maps within video codecs. Modern video codecs, like H.264 and HEVC, divide video frames into blocks, transform these blocks into coefficients, and then quantize and entropy encode them. A crucial step in this process is encoding the significance map, which indicates the locations of non-zero coefficients within a block.

The underlying idea behind ’210 is to improve the efficiency of significance map encoding by using a non-spatially-uniform partitioning of the map. Instead of assigning a unique context to each coefficient position or using uniform partitioning, the invention proposes dividing the significance map into parts where bit positions within each part share a common context. The key insight is that by carefully designing these partitions, the encoder and decoder can achieve a better balance between accuracy and adaptivity in context modeling, leading to improved compression.

The claims of ’210 focus on a method and apparatus for decoding a bitstream to reconstruct a significance map for a 4x4 transform unit. The core of the claim lies in the specific partition set used, which assigns contexts to bit positions such that the upper left quadrant has four distinct contexts, the upper right and lower left quadrants share contexts in pairs, and the lower right quadrant shares a single context. This particular arrangement is intended to optimize context-adaptive decoding.

In practice, the encoder and decoder would maintain a set of contexts, each associated with a specific partition of the significance map. For each bit position, the encoder/decoder determines the appropriate context based on the defined partition set, uses that context to encode/decode the bit, and then updates the context based on the bit's value. The non-uniform spatial allocation of contexts, with finer granularity in the upper-left quadrant, allows for more accurate probability estimation for the more frequently occurring significant coefficients.

This approach differs from prior solutions that use either a unique context for each bit position or a uniform partitioning scheme. By strategically grouping bit positions into parts with shared contexts, the invention aims to reduce the total number of contexts needed while still maintaining sufficient adaptivity. This can lead to reduced memory requirements and improved computational speed, especially in hardware implementations, without sacrificing compression efficiency. The specific 4x4 partitioning described in the claims represents a carefully chosen balance between these factors.