Video Coding Merge Candidate Generation for Bi-directional Prediction

The patent applies local quality by allowing different prediction modes (bi-directional or one-directional) to be selected for different blocks within the same picture based on local characteristics. The prediction mode decision unit determines the appropriate mode for each block individually, enabling optimal local compression without forcing a uniform approach across the entire picture, thus resolving the contradiction between temporal redundancy removal and coding efficiency.

The patent implements preliminary action by pre-organizing reference pictures into structured reference picture lists (L0 and L1) before the actual prediction process. Candidate pictures are pre-selected and arranged in display order, allowing the motion estimation unit to efficiently access relevant reference pictures without performing exhaustive searches during encoding, thus improving prediction accuracy while minimizing processing time.

The patent applies local quality by adapting the motion estimation search scope and complexity to the specific characteristics of each block. The prediction mode decision unit determines whether bi-directional or one-directional prediction is more suitable for each block, allowing motion estimation to be performed with appropriate precision and search range for each local region, thereby achieving high motion vector accuracy without uniformly high computational complexity across all blocks.