| With the development of3D display and computer hardware, the multiview viewpoint video is attracting significant interests. Multi-view stereo video includes the integration of different viewpoints, the user can experience the virtual three-dimensional scene, it also provides the viewer with the interaction between the playback system. viewers can experience different from the previous visual effects through multi-view stereoscopic video.Compared with traditional single view video, the data of multi-view video are huge and there is high correlation between inter-view sequences. Thus, how to maximize the coding efficiency of multi-view sequence is an important issue. In order to interact with viewpoint stereoscopic video, each view’s video information, disparity information and the switching information is essential, how to effectively compress the information will be the research focus in the future.At the same time, the client hardware and network environment also need to be considered. In particular, with the development of the mobile computing capabilities, watching3D TV via mobile phone is no longer a dream, the quality of mobile video services is under research. However, because the network environment is complex, how to quickly adjust current multi-view video status according to the existing bandwidth in order to improve system performance and video quality of service has become an important issue.Based on the characteristics of multiview stereo video and its high correlation between viewpoint, this paper studied several key algorithms on encoding and transmission of scalable multiview stereoscopic video.(1) We propose a novel system to estimate the depth map of a multiview video sequence, extract and match a set of scale invariant feature points. Then, we compute the constrained optical flow between selected scenes so that we can solve an over constrained linear system to estimate the depth map for each frame. This initial depth map becomes the data term of Markov Random Field formulation for depth map estimation. By minimizing the associated MRF energy function for each frame, we can refine the depth maps to achieve visually pleasing.(2) We propose a fast mode decision algorithm using both inter-view dependencies and depth information. We delicately arrange the minimal candidate modes for each macroblock according to the valued modes corresponding to its inter-view referencing macro blocks, along with the threshold based early termination to end up the mode decision procedure earlier. Next, we use the result of motion estimation to decide if disparity estimation is required and the search ranges of motion estimation and disparity estimation will be reduced accordingly. We also use depth information to refine the search range of disparity estimation and filter out the inappropriate referencing macroblocks as well.(3) We propose a GOP-level bitstream structure that completely eliminates data dependencies between any two of successive GOPs. Based on the bitstream structure, we develop GOP-level parallelization algorithms for scalable multiview stereoscopic video.(4) We propose a SMVC scheme which is designed and implemented by integrating the structures of multiview video coding and scalable video coding schemes and a layered prediction structure for the proposed SMVC scheme. The proposed scheme is designed to simultaneously support various viewpoints, various spatial resolutions, diverse picture qualities and temporal resolutions. The simulation results show that the proposed structure remarkably improves the performance of random access while maintaining very similar coding efficiency.In summary, this thesis targets the performance optimization for MVC, studies the problems of extracting depth map, mode choice, fast estimation, codec parallelization and bitstream switch, describes the problem solving process, finally gives the simulation and performance analysis results. |
PDF Full Text Request