A fast motion estimation algorithm for MPEG2 video using ripple-shaped search

Although many fast motion estimation algorithms for video coding such as H.261/3 and MPEG1 have been proposed, coding algorithms such as MPEG2 involve much larger search windows due to larger picture size and different coding structures. In this paper, we propose a novel fast motion estimation algorithm using ripple-shaped search. In the proposed algorithm, an initial, search point is determined using motion vector information of neighboring macroblocks, then a ripple-shaped search and four-step search (4SS) are used for macro and micro search, respectively. In addition, a break line method is employed to reduce the number of the mean absolute difference (MAD) calculations within a macroblock. Although 4SS can achieve almost the same speed up factors as the proposed algorithm, the PSNR performance of video coding at 4 and 10 Mbit/s by our algorithm is much better than that of a 4SS and is very close to that of full search.