Motion Compensation Prediction Using Shared Temporal Merge Candidates

The patent applies preliminary action by pre-classifying prediction blocks into two categories: those that can undergo parallel motion compensation prediction and those that cannot. This classification is performed before the actual motion compensation process, allowing the system to identify and process compatible blocks in parallel while maintaining accuracy requirements for blocks that need sequential processing. The classification criteria are established in advance based on spatial and temporal relationships between blocks.

The patent implements local quality by applying different processing strategies to different prediction blocks based on their specific characteristics. Blocks that meet certain criteria (such as having processed neighboring blocks with valid motion vectors) receive parallel processing treatment, while other blocks undergo traditional sequential processing. This localized differentiation ensures that parallel processing is applied only where it maintains prediction accuracy, thereby resolving the contradiction between productivity improvement and precision maintenance.

The patent applies partial action by selectively performing scaling operations only on motion vectors that require it, rather than uniformly scaling all motion vectors. The system identifies specific cases where scaling is necessary (when inter-frame distance differs from reference) and applies scaling only to those cases. This selective approach maintains motion compensation accuracy where needed while avoiding unnecessary computational complexity in cases where scaling is not required.

The patent implements parameter changes by dynamically adjusting the scaling factor based on the inter-frame distance and reference picture characteristics. Rather than using a fixed scaling operation, the system modifies the scaling parameters adaptively according to the specific scene and temporal distance, thereby maintaining high accuracy while optimizing computational complexity for different video content scenarios.

The patent applies merging by combining multiple small prediction blocks into larger processing units when their motion characteristics are similar. By identifying blocks with comparable motion vectors and grouping them, the system can represent multiple blocks with shared motion information using a single or reduced set of coding vectors. This merging strategy maintains the motion compensation accuracy benefits of fine-grained blocking while reducing the overall code size required to represent the motion data.

The patent implements copying by reusing motion vector information from processed neighboring blocks when applicable. Instead of encoding independent motion vectors for each block, the system copies motion information from already-processed blocks that have similar motion characteristics, particularly in regions with uniform or smoothly varying motion. This copying approach significantly reduces code size while maintaining accuracy, as the copied motion information serves as an effective predictor for similar blocks.