Research On The Key Techniques Of Multi-view Video Coding

With continuous development of the application of computer networks and the application needs, the video encoding technology is no longer just confined compression characteristics, and gradually began to the direction of network adaptability, user interaction and so on. Therefore multi-view video system which could provide stereo perception and interaction function receives more and more attention. Different from the traditional 2D video, multi-view video provides to users variety views and more realistic feel. Its applications include Free Viewpoint Video, 3D TV, Video Conferencing, Video Surveillance, Sports Games Broadcast. But compared to the traditional single view of video, multi-view video doubled the amount of data, and its storage and transmission more difficult. In addition a number of view switches between the video will increase the cost of decoder. Therefore, the multi-view video coding become the key issues of the multi-view video system applications.The paper first summarizes the development history of the multi-view coding and the research of present situation. Point out the problem needing to solve and several research hot spots. The paper focus on predictive coding technique in multi-view video coding. Takes two key technologies for the study: prediction structure and Disparity Estimation.Propose some new solutions for the problems which exist in current algorithm.(1) As a result of current prediction structure is too simple, it propose a new method which setting different prediction structure for different sequences. Experimental results show that it could obtain great effect when the sequences meet right prediction structure. It obtain better compressibility and PSNR performance compare to the Hierarchical B prediction structure.(2) Because of the prediction relationship in the Hierarchical B prediction structure is too complex which is not conducive to the characteristics of random access. It proposes a simplified prediction structure which has both coding efficiency and random access performance characteristics. Experimental results show that it has greatly improved the random access performance compare to the Hierarchical B prediction structure, and has equal quality. (3) To further enhance the multi-view video coding efficiency, it proposes a improved Global Disparity Vector technology to the problems of the current Disparity prediction algorithm. New algorithm use the characteristic that the Disparity Vector of the current macroblock has strong correlation to adjacent block to search the best disparity Vector. The experimental results show that it could save computing time by 6% while the quality of the code unchanged.