Scalable Stereoscopic Video Compression Based On H.264 MVC

Video quality has been developed from black-and-white TV to color TV, from analog TV to HDTV. The next generation of video milestone will be multi-view three-dimensional TV. When it comes to talk about stereo video transmission or storage, large volume of the original video data needs to be compressed to save bits. Therefore, it's of great value to study high-speed, high-efficiency and high-quality stereo video compression algorithm.Joint Video Team is researching on the development of the H.264 extension including Multi-view Video Coding (MVC). MVC inherits H.264's excellent features which are high-quality coding efficiency, free coding structure and good network compatibility. Besides, several new coding tools such as temporal scalable structure, viewpoint scalability, illumination compensation and hierarchical B pictures are introduced in the MVC structure. In fact, the stereo video is a special type of multi-view video. Obviously, MVC is perfectly suitable for stereo video compression. Firstly, the development history of stereo video is summarized in the first chapter. Secondly, principles of stereoscopic perception are expounded, and two mathematical models are established in order to illustrate the relationship between 3D sense and distance. Thirdly, every detail is explained about the design of stereo video player based on DavinciTM DM6446. Because of the system limitations, it's impossible to broadcast stereo video in real-time without compression. Thus, three new MVC coding structures are proposed after the study of H.264, MVC and the characteristics of stereo video. Finally, a new objective evaluation of 3D sense based on DPSNR (Disparity Peak Signal-to-Noise Ratio) is raised to evaluate the coding performance of the four structures by means of five well-designed stereo video experiments.The experiments'results indicate that, compared with the original structure, the classification performance is improved in the three new-designed coding structures. It's also proved by the rate-distortion curves that the coding performance of the third coding structure is better than the original coding structure.