Flexible Residual Transform Coding for Lower Signaling Overhead

The patent applies short distance intra prediction modes (line-by-line prediction and bi-prediction) to the residual domain before transform coding. This preliminary prediction step refines the residue blocks by exploiting local correlations, thereby reducing the magnitude and variability of the residuals that need to be encoded. By performing this prediction action beforehand, the patent reduces the number of bits required to signal the residuals without requiring complex adaptive structures during the main coding process.

The patent enables the decoder to autonomously determine transform block sizes by inferring them from the coded block size and the applied transform coding mode, without requiring explicit signaling in the bitstream. The decoder uses predefined rules and relationships between coding block sizes and transform block sizes, combined with the selected transform mode information, to self-determine the appropriate transform block dimensions. This self-service mechanism maintains transform precision while eliminating the bandwidth overhead of explicit signaling.

The patent segments the transform coding process by introducing distinct transform coding modes (e.g., horizontal transform, vertical transform, identity transform) that can be selectively applied to different residue blocks based on their characteristics. Each transform coding mode represents a segmented approach to handling different types of residual patterns. By dividing the transform coding into these manageable segments or modes, the patent achieves improved coding efficiency through mode-specific optimization while keeping the complexity of each individual mode relatively simple and manageable.