H.264 Algorithm Based On Motion Vector Difference CABAC Improvements

In the H.264/AVC video compression standard, context-based adaptive binary arithmetic encoding (CABAC) is an efficient entropy coding algorithm which contains the binaryzation of syntactic elements, the context model selection of bit characters, the probability updating of bit characters, the initialization of this encoding algorithm, the binary arithmetic coding and so on. Meanwhile, as an important syntax element of CABAC coding theory in H.264/AVC video compression standard, motion vector difference (MVD) is the focus of the study in the thesis which includes main research and innovations as following.(1) For the large block partition modes such as16×16,16×8,8×16&8×8in the inter coding frames (P frames and B frames), due to the correlation between the current coding block and adjacent encoded blocks is low and it can’t effectively predict MVD values of the current coding block by using the adjacent encoded blocks, an approach that can improve the forecasts accuracy of MVD of the large block partitions is proposed based on the block partitions modes.(2) For the features that the numbers of large block partitions contain the most ones in inter coding frames, which demonstrate the numbers of MVD that corresponding to the large block partitions occupy the majority ones, and MVD coding bits are the one of most coding ones in CABAC encoding, so it can confirm that the improvement about MVD coding efficiency can enhance efficiency of CABAC encoding as a whole.(3) In the two components of the MVD, the conditional probability distribution between|mvd-y|and|mvd(A,y)|+|mvd(B,y)|are very similar to these between|mvd-y|and2×|mvd-x|. Therefore, it can effectively improve the accuracy of the first bit of the MVD vertical component by using it in context model selection as well as save the numbers of coded bits in the inter predictive coding frame.(4) For characteristic that, in the context model selection, the first bit of MVD’s each component in the current coding block is not only related to neighboring encoded blocks but has a very high correlation with the MVD’s corresponding component in the current coding block, an optimization algorithm that based on the consideration of various correlation is proposed. By using it, it can effectively improve the accuracy of the first bit of the MVD components in context model selection.Due to the standard CABAC algorithm fail to make a more effective choice for the MVD components’first bits in the context model selection, it will make those bits fail to make a more balanced choice of the context models for arithmetic coding, which will inevitably increase the times of normalized and computational complexity in the whole coding process. Therefore, based on the block partition modes, an optimization algorithm (CABAC-1algorithm) is proposed in the paper by taking advantage of the relationship between mvd-y and mvd-x and the correlation between first bits of each MVD component and their own corresponding MVD components. The experiments show that, compared with the standard CAB AC algorithm, the CAB AC-1algorithm can ensure the quality of the encoded video as well as save about8%bit numbers of P frame and save about5%bit numbers of B frame. Meanwhile, it can reduce about2%coding time. Consequently, the CAB AC-1algorithm is an effective improved algorithm.