Research Of Motion Estimation Technology In H.264/AVC

In recent years, with the rapid development of multimedia technology, the latestvideo coding standard H.264/AVC also came into being. H.264/AVC video codingstandard is developed by Joint Video Team which is composed of Video CodingExperts Group of International Telecommunication Union TelecommunicationStandardization Organization and Dynamic Picture Experts Group of ISO/IEC.Because of this new video coding standard including many new technologies, it has ahigher coding efficiency compared with the conventional coding standards. However,at the same time it greatly increases the complexity of the encoding. Motionestimation is an important part of the whole video compression coding system, whosetime cost accounts for about80%of the entire coding time. Therefore, the research offast motion estimation algorithm to improve the efficiency of the motion estimation,to saving the encoding time and improve the real-time has great significance.This paper first introduces the basic principles of block-matching-based motionestimation and some classic motion estimation algorithm, such as full searchalgorithm, three-step search algorithm, diamond search algorithm and hexagonalsearch algorithm. The paper goes on to introduce the fractional pixel accuracy motionestimation as well as fractional pixel accuracy motion vector characteristics. Then,this paper further studies the UMHexagonS which is used by JM codec referencemodel of H.264/AVC video coding standard. On this basis, the paper dose someoptimization and improvement on the aspects of dynamic search range estimationalgorithm, non-symmetrical cross search,5×5rectangle search and the hexagonsearch of non-uniform multi-level. After that, this paper proposes a new fractionalpixel motion estimation algorithm based on the highly correlation according to thefractional pixel motion vector and the CBFPS.Based on some test sequences, some experiments were done on the codecreference model JM11.0of H.264/AVC video coding standard in order to compare theUMHexagonS the algorithm and CBFPS algorithm with the new algorithm proposedin this paper. As can be seen from the experimental results, the whole pixel and fractional pixel motion estimation algorithm proposed in this paper, have differentdegrees of reduction in the premise of essentially the same image quality and bit-rate.