Research On Inter Prediction Motion Estimation Based On H.264

H.264/AVC is an video encoding standard proposed together by ISO/IEC MEPG (Moving Picture Experts Group) and ITU-T VCEG(Video Coding Experts Group) in 2003.In May, it formally established as an international standard. The standard has the advantage of a high compression ratio, strong network adaptability, low bit rate, etc.H.264/AVC uses multiple reference frame selection, advanced intra prediction and inter prediction precision, high-precision motion estimation and other new technologies. It greatly enhances the coding efficiency, but the complexity of the algorithm also will greatly be enhanced.In various coding techniques of H.264/AVC standard, motion estimation technology is the core technology, for eliminating the temporal redundancy in video sequences. Motion estimation algorithm takes up more than 70 percent of the total encoding time, and its coding efficiency determines the efficiency of the entire video encoding, so fast motion estimation algorithm has become the focus of research in the field.This thesis introduced the basic concepts and core principles of H.264/AVC standard. It also analyzed key technologies of the H.264/AVC standard, especially motion estimation technique. Then it introduced several classical algorithms which based on block matching motion estimation. Then, it verified the effectiveness of an improved EPZS (Enhanced Predictive Zonal Search) algorithm. This thesis focused on UMHexagonS (Unsymmetrical-cross Multi-Hexagon-grid Search) algorithm and detailed the basic steps. Based on the idea of reducing the search point, the algorithm improved from three aspects. Firstly, the early termination of the initial search point prediction; secondly, adaptive judge movement type; At last, hybrid search template improvements. Using different video sequences, the optimized algorithm has been tested with the original comparison from encoding time and image peak signal to noise ratio. The optimized algorithm saved 19.55 percent of encoding time and increased 0.08dB of the average of PSNR. The image on the subjective almost had no change, thus proving effective for optimizing the original algorithm.