Research On Fast Algorithm Techniques For Multi-view Video Coding Based On H.264

With Free-viewpoint Television and 3D Television developments, MVC(Multi-view Video Coding) has received critical attention from academia as critical technology in recent years. Multi-view stereo videos contain a large amount of data, because of that, MVC adopt complex prediction structure and various prediction technology. Meanwhile, high compression efficiency is achieved at the expense of substantially increased computational complexity. Various prediction modes and inefficient disparity estimation are the bottleneck of coding speed. This paper propose a fast algorithm for MVC mode decision and disparity estimation, including:1. A fast macroblock mode decision algorithm based on depth maps is proposed. By analyzing the statistical characteristics of the best mode and distribution of modes under different depth, then selecting corresponding mode search strategy for each macroblock according to the characteristic of space region. Near region, Middle region and Far region can be extracted from space region with depth maps. We present a method of Far region segmentation based on morphology. In this region, macroblock will be encoded as SKIP, Interl6×16 modes and limited to intra-view. We also present a method of Near/Middle region segmentation that is combination of background subtraction and frame difference. In Near region, macroblock will be encoded as SKIP, Inter16×16 and Intra modes. Experimental results show that our method can achieve 53.14% of time saving on average.2. To solve the inefficiency of global disparity estimation based on SIFT or gray match, a improved algorithm for global disparity vector estimation based ORB(Oriented FAST and Rotated Brief) is proposed. ORB image feature can be used to match images in different views and calculate global disparity vector estimation. ORB employ FAST corner detection which is a simple and efficient algorithm to detect keypoint., and match those descriptors based on hamming distance. Experimental results show that the global disparity vector extraction method we propose can speed up about 5 times compared with SIFT.3. By analyzing the role of 3D Warping technology in disparity vector prediction, a united algorithm based on global disparity vector and 3D Warping technology is presented for optimizing searching starting point for disparity estimation and predicting direction of disparity vector. The prediction of direction is then used to optimizes searching range. By analyzing the relationship between disparitys and depth, searching starting points and searching range can be further optimized based on depth information. Experimental results show that the proposed fast disparity estimation algorithm can reduce encoding time by more than 40.54% while the quality of coded image are not compromised.This paper exploited depth information to decide coding mode, speed up disparity estimation combining various prediction technique, adopt ORB to accurate computing global disparity vector. The experiment results prove the algorithm we described enhance coding speed and has greater application value for multi-view video plus depth video format.