Research And Optimization On The Motion Estimation Algorithm In H.264

H.264is the newest video coding standard. Compared with the last video coding standards, h.264adopts some new technology, such as Multiple reference picture motion compensation, Variable block-size motion compensation, Quarter-sample-accurate motion compensation and Weighted prediction. So h.264achieves better coding efficiency, but it also has high computational complexity. Motion estimation is very important in the video coding standard, and it is the most time-consuming part. So the efficiency of the coding system is decided by the efficiency of motion estimation. It is necessary to research and optimize motion estimation algorithm.This paper particularly discusses the basic principles and key technics of h.264video coding standard. On deeper study on the current developing of the motion estimation algorithms, this paper introduces several motion estimation optimization strategies and their related typical fast motion estimation algorithms. Their merits and demerits are also analyzed.This paper gives detailed introduction of UMHexagonS algorithms, and analyses the algorithm’s three problems. To solve those problems, this paper proposes an improvement on UMHexagonS.Three dynamic models are used in early termination threshold, search range and search mode respectively, which enhance the algorithm’s adaptive ability. Six video sequences with different motion levels are tested in the experiment. Comparing to UMHexagoS algorithm, the proposed algorithm can save21.67%encoding time and47.94%motion estimation time on average, with only0.02dB degradation in the peak signal-to-noise ratio (PSNR) and1.69%increase in the bit rate.According to the overall considerations of various motion estimation optimization strategy, this paper proposed a fast motion estimation search algorithm using dynamic modes. Calculating motion magnitude and direction of macro-blocks, the proposed algorithm chose a corresponding search mode for them, at the same time, the proposed algorithm had an improvement on the media prediction in the standard and proposed an dynamic strategy that enabled the reference block stop early. The experiment shows that the proposed algorithm can decrease motion estimation time while keeping the same rate-distortion performance. Compared with FFS and UMHexagonS, the proposed algorithm can save85.28%and35.29%motion estimation time respectively.