Context-based Adaptive Binary Arithmetic Coding Efficiency Improving For H.264

Context-Based Adaptive Binary Arithmetic Coding (CABAC) is entropy coding in H.264/AVC standard. The compression ratio of CABAC has improved 30% compared with Context-Based Adaptive Variable Length Coding (CAVLC) and the other traditional VLC. But the coding context modeling, probability estimation, and renormalization process have greatly increased the computational complexity.Based on the basic theory of entropy coding, this thesis improves and optimizes the CABAC entropy coding algorithm on three aspects, by applying improved methods of CABAC at home and abroad.First, the thesis describes the improvement of the MVD (Motion Vector Data) component prediction Model. The original algorithm which used the encoded MVD neighboring block information probability model can not achieve a high coding efficiency. Using a horizontal component of the current encoded macro block to predict the vertical component of the current macro block is proposed in this thesis. The method is based on different macro block sizes, through three different conditions on the macro block to predict the vertical component, and improve the accuracy of prediction. Experiment results indicate that the proposed algorithm improves 8%(P frames MVDy) and 5%(B frames MVDy) coding efficiency.Then the thesis optimizes the renormalization process. In order to reduce the coding time, this thesis proposes a fast renormalization algorithm based on interval division to deal with the "bottleneck" problem in renormalization. According to the number of cycles in renormalization loop, this fast renormalization algorithm introduces six different interval division standards to replace time-consuming, bitwise-operation procedure. At the same time, the algorithm can outputs more bits to coding stream in unit time and better meets the requirements of real-time, by removing the cycle of renormalization procedure. Experiment results indicate that compared with original renormalization algorithm, the reduction of renormalization algorithm number and their corresponding run time are 21.9%-26.7% and 14.5%-33.7% respectively. In the last part, the thesis optimizes the bypass coding algorithm. In order to reduce the coding time, a fast bypass coding algorithm based on code model is proposed. The algorithm proposes different encode models to accelerate the speed of the bypass encoder based on the length of coding. Experiment results indicate that the proposed bypass coding algorithm enables reduce the number of bypass coding algorithm by 18.9% to 30.5% and their corresponding run time by 32.9%～38.9%.