Square-diamond search algorithm for block motion estimation

This paper firstly analyzes some typical search algorithms in motion estimation, especially the Diamond Search(DS)algorithm. It is found that these algorithms are all based on serial processing ideas, since their search steps can only be changed degressively, in other words, first coarse location and second accurate orientation, which make the search blindness. Secondly, aimed at questions existing in these algorithms, the authors design a new integrative pattern-Square Diamond Pattern (SDP), which is composed of a diamond and a square. The SDP can realize content-based search for following three possibilities when it performs matching computation. If the minimum block distortion (MBD) point in the middle of the pattern, it shows that image is still and completes search by one step, if the MBD point is at one of the diamond's four corners, it shows there is small motion in the image, if the MBD point is at one of the square's four corners, it shows there is large motion in the image. The next step of search will adaptively use different pattern according to motion types. Therefore, SDP is based on parallel processing idea of coarse location and accurate orientation determined by its structure. Thirdly, authors present a new Square-Diamond Search (SDS) Algorithm for block matching motion estimation with SDP. Finally, the results of experiments show that not only the new SDS is much faster than traditional algorithms, but also its PSNR and visual quality of the retrieval images are better than those of other algorithms, and as nearly good as that of FS.