| According to the data, people capture about 70% percent information comingfrom vision, so video information is very important in multimedia: at the same time,video data has much redundancy, the key factor of multimedia's service quality is thevideo quality after compression. The key technology of multimedia application isDigital Video. That is to say the key technology of the use of multimedia is solvingthe conflict of the data and the communications networks, and the solution of whichis compression. So the encoding of video is the key technology of information field.For the enhancement of the coding efficiencies, the new standard adopts newtools as followed: multiple reference picture, variable block-size with seven blocksizes in motion prediction, quarter-pixel accuracy for motion vector, short wordlengthinteger transform and loop filter deblocking.At the same time, the complexity of the H.264 codec is several times higher thanthat of existing standards(MPEG-2 and H.263). Accordingly, the software-based realtimevideo codec demands higher-powered processor and faster-speed algorithms.Due to this cause, many researchers tried their best to study the H.264 encoder tospeed up it. However, they mainly concentrated on the motion estimation algorithmoptimization and mode decision algorithm optimization. But the algorithm this paperproposes is motion estimation algorithm.This thesis introduces video en/decode standards and their principles, anddescribes the new coding feature of the H.264 and the basic knowledge of MotionEstimation. After investigating scores of algorithms, this article has summarizedsome of the algorithms completely and introduces an improved ME search algorithm.The improved ME search algorithm is on the base of algorithm UMHexagonS. Firstly,improved the search range in the circle c. Secondly, the 5×5 full-search is replacedby an octagon search and a small diamond searches. Thirdly, used early terminationin the hexagon search and the extended hexagon search. The simulation results whichthe improved UMHexagonS algorithm runs in JM 10.1 model, show that theimprovement can reduce 50~70%period of motion estimation under the premise oflittle change on PSNR, compared with those of the original UMHexagonS algorithm. |