Researches On The Compression Algorithm Of Multiview Video Coding For 3D Video

With the rapid development of multimedia technologies and higher user demands, three dimensional (3D) video becomes a trend in the consumption area, attracting a great attention.3D video, compared to traditional two dimensional (2D) video, can provide more realistic stereoscopic experience. It commonly consists of multiview texture video sequences and corresponding depth information. The amount of the data is large, which is a challenge to storage and transmission especially under limited bandwidth. Thus it is necessary to efficiently compress and transmit the 3D video data, particularly the multiview texture information which constitutes the largest portion. And the key to promote the use of 3D video lies in the efficient coding and transmission of multiview video. In conclusion, it is of great theoretical and practical significance to do research on the compression of multiview video.This thesis focuses on the key technologies of multiview video coding for 3D video applications. And the main work and innovation are as follows.Firstly, this thesis proposes a novel scheme that enables viewpoint switching in multiview video streaming. In a typical free viewpoint video streaming system, users'demands are different. It is obviously inefficient to transmit the entire multiview stream to every different client, and viewpoint switching methods are usually adopted. We present a new scheme that enables viewpoint switching through inter-view-predicted redundant pictures. For each desired switching point, sets of redundant pictures are encoded using only inter-view prediction. When switching happens at a switch point, an appropriate redundant picture is adaptively chosen based on which views were transmitted before and will be transmitted after the switch point, and this redundant picture is transmitted instead of the original one to enable efficient viewpoint switching. Experimental results show that the proposed viewpoint switching method not only efficiently saves the transmission bandwidth but also improves the compression performance.Secondly, the thesis investigates the impact of introducing view synthesis prediction into multiview video coding on the overall performance of 3D video coding. In 3D video applications, whether 3D video displaying or free-view browsing, user terminals make use of the depth information just for synthesizing intermediate virtual views not for displaying. Hence it is very meaningful to try to reuse the encoded depth maps available both at the encoder and the decoder to improve the coding efficiency of the 3D video. In this thesis, a large number of experiments have been done to investigate the impact of view synthesis prediction on 3D video coding. The results demonstrated that VSP will improve the performance of multiview coding. Moreover, it shows that the synthesized virtual view will provide more accurate prediction than those existing reference views. Additionally, our research provides large amounts of accurate experimental data for the researchers in the fields concerned.