A New Multi-hypothesis Motion-compensated Prediction Algorithm
一种新的多假设运动补偿预测算法

Multi-hypothesis motion compensated prediction has been used as bi-directional prediction mode in B picture of the H.264/AVC video compression standard.The optimal motion-compensated prediction signal,named hypotheses,is composed of blocks in both forward and backward reference pictures,which are simply added and averaged to synthesize the final prediction signal.In the original algorithm,more than one motion vector and reference information has to be encoded and the accuracy of motion-compensated prediction is still insufficient.To overcome these disadvantages,a new rate-distortion optimization based MHMCP algorithm is proposed.After the optimal MB's encoding mode has been obtained by motion estimation and mode decision procedure,bi-directional motion search with iterative local motion vector refinements will be started.At the same time,two extra mv-tracking modes which need not to encode additional side information will be examined,accompany with adaptive hypothesis-coefficients adjustment.Experimental results show that not only the energy of the residual signal can be decreased but also the quality of the motion-compensated prediction signal can be improved.Therefore better R-D performance can be achieved by the proposed algorithm.