H.264 Based Fast Motion Estimation Technique

H.264 is video coding standard which has good compression performance but expensive computation. Block based motion estimation (ME) is the most time-consuming component in an H.264 video encoder. The ME process consists of two stages:integer pixel search and fractional pixel search. H.264 standard has adopted Unsymmetrical cross Multi Hexagon grid Search (UMHexagonS) algorithm for integer pel motion estimation and Center Biased Fractional Pel Search (CBFPS) algorithm for fractional pel motion estimation. UMHexagonS and CBFPS algorithms show very good capability in keeping the rate distortion performance, as well as achieve a great computation reduction. However, UMHexagonS algorithm needs large amount of computing in integer pel motion estimation. There are still some weaknesses in threshold and the starting point searching in CBFPS algorithm. In this paper, these problems have been studied as following:(1) In this paper, we propose an improved integer pixl motion estimation algorithm to solve the insufficiency of the original UMHexagonS algorithm. The algorithm improved the prediction of the original search center, and made a change for the search pattern of UMHexagonS, search layer, and application range of 5 X 5 search method. These changes reduced search points greatly of the integer pel motion estimation, and enhanced the efficience of motion estimation.(2) Mearnwhile, to reduce the complexity of fractional pel ME, we propose an enhanced fast franctional motion estimaton algorithm. The algorithms reduce search points by skipping to search the blocks which has been tested as no movement block. Futhermore, we propose an algorithm named MSDSP to reduce the caculation in searching. The algorithms improve the fractional search pattern by mixing small diamond search pattern which can avoid Local optimum.The proposed algorithm can reduce more than 20% search points with minimal loss in bitrate and reconstructed quality compared with those of the original UMHexagonS algorithm. Compared to the CBFPS sub-pel ME algorithm proposed in H.264, the proposed fractional pel ME method can reduce more than 70% of fractional pixel search points respectively, with a negligible degradation in quality. The methods in this paper can be applied in real-time video encoding with good efficiency and robustness.