Predictive block-matching discrepancy based rhombus pattern search for block motion estimation

In this paper, we propose a predictive block-matching discrepancy based rhombus pattern search algorithm for block motion estimation. The block-matching discrepancy is formulated as sum of absolute difference (SAD). Due to the highly correlated block-matching SADs of adjacent blocks in the same frame, the SADs of neighboring blocks are used to predict the SAD of current block. Based on the predicted SAD, we design an early decision mechanism by using an adaptive threshold to early terminate the search if the SAD of the search point is smaller than the adaptive threshold. Furthermore, to decide the search priority of the points in motion estimation, we propose a pattern search method, which consists of the selection of the starting points and a rhombus pattern search. Accordingly, a predictive SAD based rhombus pattern search algorithm, named PS-RPS, is proposed. The experiment results demonstrate that with similar video quality in peak signal-to-noise ratio (PSNR), the average search speed of PS-RPS is 499 times faster than that of full search (FS), 7.66 times faster than that of the diamond search (DS), and 2.2 times faster than that of the adaptive rood pattern search with zero-motion prejudgement (ARPS-ZMP).