Motion Refinement Engine Shared Memory Architecture

The patent applies preliminary action by pre-calculating and storing neighbor data for each macroblock in a shared memory structure before the actual motion compensation process. This preprocessing step allows the motion search and refinement operations to proceed without real-time neighbor data retrieval delays, thus maintaining encoding accuracy while reducing processing time.

The patent segments the video encoding process into distinct stages: neighbor data preprocessing, motion search, and motion refinement. By dividing the macroblock processing into independent segments that can be handled separately and stored in shared memory, the system achieves better parallelization and reduces the critical path delay, improving encoding speed without sacrificing accuracy.

The patent merges the neighbor data management function into the motion compensation unit by implementing a shared memory structure that is jointly accessed by both motion search and refinement operations. This consolidation eliminates redundant neighbor data retrievals and allows parallel processing of multiple macroblocks, thereby increasing encoding speed while maintaining manageable complexity through unified data access protocols.

The patent introduces a shared memory structure as an intermediary between the neighbor data generation and the motion compensation operations. This intermediary buffer stores pre-computed neighbor data that can be efficiently accessed by multiple macroblock processing units simultaneously, enabling parallel execution without increasing the complexity of individual data management paths.