Video Coding Efficiency via Multiple Hypothesis Motion Candidates

The patent segments the motion prediction process into multiple hypotheses, where each hypothesis represents a different prediction approach (e.g., temporal prediction, spatial prediction, combined prediction). By dividing the prediction task into multiple segments and evaluating each, the system achieves better coding efficiency without overwhelming computational burden.

The patent applies partial action by selectively evaluating multiple hypotheses only when beneficial, rather than always using all prediction methods. The system determines whether to use multiple hypotheses based on rate-distortion optimization, applying the additional computational effort only when it improves coding efficiency beyond a certain threshold.

The patent performs preliminary action by pre-calculating and storing motion candidates in the history-based motion candidate table before actual prediction is needed. Multiple hypotheses are evaluated in advance, and the best candidates are stored for rapid retrieval during decoding, reducing real-time computational complexity.

The patent uses copying by creating multiple hypothesis copies of motion information from reference blocks. Instead of performing complex real-time calculations, the system copies motion vectors and prediction data from previously decoded blocks, storing multiple versions (hypotheses) in the candidate table for quick comparison and selection.