Study On The Binarization Scheme Of Motion Vector Difference In CABAC

In H.264/AVC, two methods of entropy coding are adopted, context adaptive variable length coding (CAVLC)and context adaptive binary arithmetic coding (CABAC). CABAC is a highly efficient entropy coding method. Comparing to CAVLC, CABAC can significantly improve the coding efficiency of H.264/AVC. CABAC can be divided into three parts, binarization, context modeling and binary arithmetic coding.Coding the motion vector difference (MVD) in CABAC occupy a large part of bits. For MVD in H.264/AVC, UEGk binarization is employed with unified binary parameters. Actually, the MVD probability distribution is different under different quantization parameter (QP). The current binarization scheme in H.264/AVC neglects the difference between the probability distribution, can not provide a good approximation of the true probability distribution of MVD. Motivated by this feature, different sets of parameters for MVD's UEGk binarization under different QP is proposed in this dissertation. Experimental results show that the bits for predictive coding frames (P-frame) can be decreased by using the proposed scheme and the coding efficiency can be slightly improved. Based on the same principle, adjusting the binarization parameters for residual et al should improve the coding efficiency too.