Analysis and optimization of the UMHexagonS algorithm in H.264 based on SIMD

Motion estimation is the part of the most computational complexity and the greatest impact on the performance in H.264. The UMHexagonS is one of the main motion estimation algorithms of H.264, whose with features of high computational complexity and time-consuming. The main computation load of UMHexagonS comes from sum of absolute differences of block-matching operations. Due to pixels´ relevance of sum of absolute differences is not exists, so an optimization method of SIMD can be used for sum of absolute differences of block-matching operations. On the base of analysis of UMHexagonS algorithm, an optimization based on JM 12.2 encoder was made by SIMD execution model on IA-32 architecture. Experiment results shown that the optimization can effectively enhance coding speed without any losing of PSNR.