Research On Side Information Generation And Fusion Algorithm For Multi-view Distributed Video Coding

Multi-view video coding can provide wealth video scenes information, which receives more and more attention. But multi-view video coding adds a lot of redundant information and increases the complexity of system. One solution to this problem is to use a multi-view distributed video coding.The core algorithm of multi-view distributed video coding is side information generation and fusion algorithms. The higher the quality side information, the less parity information delivery from encoding to the decoding, and the higher the reconstruction quality is. In a single viewpoint, the distributed video coding algorithms using only generated temporal side information by the temporal correlation. While in multi-view distributed video coding algorithms, they include not only the temporal side information, but also generated inter-view side information using the view correlation. The two side information through fusion algorithm can effectively reconstruct the video information. This paper studies distributed multi-view video coding side information generation and fusion algorithms, specifically including the following aspects:Firstly, as for the problems of the existing temporal side information generated at decoding, it not including the frame information and it not global, this paper proposes a novel side information generation and fusion algorithms. The global side information of this algorithm is extracted by the SURF algorithm at the encoding side. The inter-view side information, the global side information, the local side information can obtain fusion side information through the fusion function algorithm to reconstruct the non-key frames. This method can effectively improve the quality of the reconstructed frame. Secondly, as for multi-view distributed video coding mainly uses the motion compensated interpolation to reduce the temporal redundancies inside the viewpoint. The decoding order of motion compensated interpolation is inconsistent with the video sequence, and the video decoder appears to decoder delay. An efficient low-delay multi-view distributed video coding is proposed. The experimental results show that the presented method can reduce system delay in the case of considerable reconstruction quality. Finally, as for different number of viewpoints, the distribution of different viewpoints position, this paper research the applicability of disparity vector interpolation and extrapolation in multi-view distributed video system, and prove this in theory.