Video Decoder Merge Candidate List Construction

The patent extends the spatial search from adjacent blocks (2D neighboring positions) to include non-adjacent diagonal blocks by incorporating temporal dimension information. This allows candidates from previous and future frames to be included in the merge candidate list, effectively adding a temporal dimension to the spatial search and increasing the pool of available candidates from limited adjacent blocks to multiple temporal layers.

The patent performs preliminary construction of merge candidate lists by pre-identifying and storing motion vectors from temporal reference frames before the actual encoding/decoding process. This preliminary action ensures that when adjacent spatial candidates are insufficient, pre-computed temporal candidates are already available to immediately supplement the list, avoiding compression efficiency loss.

The patent implements a dynamic merge candidate list construction process where the number and type of candidates are adaptively adjusted based on availability. The system dynamically selects from spatial, temporal, and combined candidates, and dynamically supplements the list when candidates are insufficient, rather than using a fixed large number of candidates. This dynamic approach maintains high prediction accuracy while avoiding unnecessary computational complexity from always processing maximum candidates.

The patent applies different candidate selection strategies to different regions and scenarios. For blocks with sufficient adjacent spatial candidates, traditional methods are used. For blocks with insufficient candidates, temporal candidates are selectively added. This localized quality approach ensures computational resources are focused only where needed, improving prediction accuracy for problematic blocks without increasing overall computational complexity for all blocks.

The patent implements a self-service mechanism where the merge candidate list automatically supplements itself with temporal candidates when spatial candidates are insufficient. The system monitors the number of available candidates and autonomously adds temporal motion vectors from reference frames when the count falls below thresholds, eliminating the need for external intervention or placeholder additions. This self-service approach ensures list completeness with meaningful candidates rather than duplicates or zeros.

The patent introduces temporal reference frames as intermediary sources to bridge the gap between insufficient spatial candidates and the required minimum candidate list. Instead of adding duplicate or zero candidates directly, the system uses temporal motion vectors from previous and future frames as intermediary candidates, providing valid prediction options that maintain list completeness without degrading compression efficiency through unnecessary computations.