Spatial-Temporal Context Derivation for Inter Prediction Coding

The patent changes the parameters used for context determination from traditional methods to a new approach based on motion vector prediction parameters and spatial neighboring block information. This involves deriving parameters such as motion vector differences and reference frame indicators to select coding contexts, thereby improving compression efficiency while managing complexity through structured parameter derivation.

The patent applies partial action by using context-adaptive binary arithmetic coding (CABAC) with selectively chosen contexts for coding motion vector syntax elements. Instead of uniformly allocating bits to all motion vector components, the method selectively applies precision where needed based on derived parameters from spatial neighboring blocks, thereby improving decoding accuracy for critical elements while controlling overall bit rate.

The patent segments the context determination process into distinct stages: first deriving parameters from spatial neighboring blocks (such as motion vector differences and reference frame indices), then using these parameters to select appropriate coding contexts. This segmentation allows the system to manage processing complexity by breaking down the optimization task into manageable steps while maintaining coding efficiency gains.