A Fast Motion Estimation Algorithm Based on Optimized Sampling

A fast motion estimation algorithm based on global sampling is presented in this paper. It runs global sampling search with multiple precision in the search window first and then runs 2-step fine search with multiple patterns, which significantly avoids the search falling into the local-minimum point. Making use of the temporal and spatial features in video sequences, the proposed algorithm optimizes the search process and employs new prediction and early-termination strategies to reduce its computational complexity and improve its search accuracy. Experimental results show that, compared with UMHexagonS algorithm in different kinds of test sequences including HD format, the proposed method averagely reduces 32% of motion estimation time with almost the same R-D performance and bit rate.