Multi-hypothesis Inter Prediction Bandwidth Reduction

The patent extracts only the essential motion information from multiple hypotheses rather than transmitting complete motion data for each hypothesis. By selecting and transmitting only critical motion parameters (such as motion vector differences or selected hypothesis indicators) instead of full motion information, the bandwidth requirement is reduced while maintaining the prediction efficiency benefits of multi-hypothesis mode.

The patent applies partial action by using a subset of the full multi-hypothesis information set. Instead of utilizing all available hypothesis data which would maximize bandwidth consumption, the method selectively uses only the necessary portion of motion information from multiple hypotheses to achieve sufficient prediction accuracy, thereby reducing transmitted data while maintaining improved prediction efficiency.

The patent performs preliminary action by pre-generating and caching motion candidates during the encoding process before actual prediction is needed. Motion vectors and candidate information are prepared in advance and stored in candidate lists, allowing the decoder to benefit from multiple hypotheses without performing computationally intensive motion estimation at decode time. This preliminary preparation improves coding performance while managing computational complexity through advance computation.

The patent applies dynamics by adaptively adjusting the number and type of motion candidates generated based on picture characteristics, block size, and prediction mode. The system dynamically modifies the complexity of multi-hypothesis processing - using more candidates for difficult prediction regions and fewer candidates for simpler regions - thereby improving coding performance where needed while reducing overall computational complexity through selective application.