Geometric Partitioning and Merge Reordering for Video Coding

The patent segments the set of geometric partitioning modes into multiple groups, where each group contains a subset of partition modes. Instead of signaling all possible partition modes, the encoder only needs to signal the group index and the selected mode index within that group, significantly reducing the number of bits required for mode signaling while maintaining the ability to represent all geometric partitioning configurations.

Different groups of partition modes are assigned to different local regions or contexts based on their characteristics. The patent applies different partition mode groups to different coding units or prediction units depending on local motion characteristics, block size, and other contextual factors, optimizing the balance between coding efficiency and signaling overhead for each local region.

The patent performs preliminary computation of template matching costs or boundary matching costs for all merge candidates before the actual encoding process. These costs are pre-calculated and stored, allowing the encoder to quickly reorder merge candidates based on pre-computed metrics rather than calculating them on-the-fly during encoding, thus reducing real-time computational complexity while maintaining improved encoding performance.

The merge candidate reordering process uses self-generated cost information (template matching costs or boundary matching costs) to automatically determine the optimal ordering without requiring external guidance or iterative optimization. Each merge candidate's position in the reordered list is determined by its own pre-computed cost metric, enabling efficient self-organization of the candidate list.