DS算法.pdf

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 2, FEBRUARY 2000 287 A New Diamond Search Algorithm for Fast Block-Matching II. OBSERVATIONS Motion Estimation Fundamentally speaking, the search patterns shape and size ex- Shan Zhu and Kai-Kuang Ma ploited in the fast algorithm jointly determine not only its search speed but also resulted performance. Block distortions (or block-matching errors) form an error surface over the search window, and the global Abstract—Based on the study of motion vector distribution from sev- minimum point corresponds to the motion vector where the best eral commonly used test image sequences, a new diamond search (DS) al- matching (or the least error) incurs. Since the error surface is usually gorithm for fast block-matching motion estimation (BMME) is proposed not monotonic, multiple local minimum points generally exist in in this paper. Simulation results demonstrate that the proposed DS algo- rithm greatly outperforms the well-known three-step search (TSS) algo- the search window especially for those image sequences with large rithm. Compared with the new three-step search (NTSS) algorithm, the DS motion content. Therefore, searching with a small search pattern, such algorithm achieves close performance but requires less computation by up as the one used in BBGDS [8] with size of 3  3, is quite likely to be to 22% on aver