Inter Prediction Merge Estimation Parallel Processing

The patent divides the picture into multiple independent merge estimation regions (MERs), where each MER can construct its merge candidate list independently without relying on motion data from other regions. This segmentation breaks the data dependencies that prevent parallel processing, allowing multiple PUs within the same MER to be processed simultaneously while maintaining motion estimation accuracy through proper candidate selection within each region.

The patent applies different merge candidate construction strategies to different regions based on their specific characteristics. Within each MER, the patent selectively includes spatial candidates from adjacent PUs that are available in parallel processing order, while excluding candidates that would create dependencies. This local adaptation allows each region to have optimal candidate availability tailored to its position and the parallel processing constraints, rather than applying a uniform restriction across the entire picture.

The patent establishes a predetermined parallel processing order for PUs and pre-determines which spatial candidates will be available at each processing step. By planning the candidate selection in advance based on the processing order, the patent ensures that only PUs with guaranteed available motion data are included as candidates, eliminating the need for runtime availability checks and ensuring reliable parallel processing execution.