Research On Predictionand Post Process Technology Of Multi-view Video Coding

As the developing direction of the next generation multimedia application,multi-view video meets the freedom to choose the viewing angle by providingmultiple view points of video data. It has a three-dimensional and interactiveoperations function. Its amount of data increases with the number of cameras growslinearly. How to effectively compression the data becomes the key issue of it’s widelyuse. As one of the key technologies to achiece freedom of choice viewpoint, virtualview synthesis uses the adjacent color views and depth maps to generate the virtualview, the quality of depth map determaines the quality of virtual view largely.The data of multi-view video has a strong correlation between views. The JointVideo Team expands the H.264/AVC video coding standard to support multi-viewvideo coding and uses the disparity estimation to eliminate redundancy amontviewpoints. But the disparity estimation methods are not very efficient because of theocclusion and deformation between different viewpoints. For the depth map, theexisting compression coding methods are used but add a post-processing step toimprove the quality of virtual view synthesis. This paper focuses on a study of theseissues, proposed a inter-view prediction method based on low-rank matrix recoveryand a edge-adaptive depth map post-processing method. The main research focusingon the following aspects:1. This paper proposes an inter-view prediction method based on low-rank matrixtheory for the redundancy between viewpoints and the occlusion and deformation.From the color view and depth map of adjacent viewpoints we can get the virtualview of the current viewpoint. Using the correspongding view as reference we canfind several similar blocks of the current block to construct a low-rank matrix. Thenwe use low-rank matrix recovery method to remove the noise signals of the similarblocks, and use the result as prediction block. The experimental results show that thismethod can effectively improve the efficiency of inter-view prediction.2. For the depth map post-processing, this paper presents an edge-adaptive processmethod. Firstly get the depth map and color map edge by edge detection algorithm.And then explicit the smooth region, the edge and the discontinuous caused bycompression by align the edges. And use different filtering methods to process theregions. The experimental results show that this method can obviously improve bothobjective and subjective qualities of vitual view synthesis.