Research On Technology Of Depth Map Compression In3DV System

As a new type of digital media, multi-view video make users have the freedom tochoose the point from various angles to watch3D scene by providing video datas ofmultiple viewpoints, whose unique stereo feeling and interaction capability betweenviewpoints make it widely used in3DTV, FTV and other fields. Multi-view videocontains a large number of datas, which leads to the huge pressure of storage andtransmission. Therefore, how to compress multi-view video efficiently becomes thebottleneck of restricting its development. Depth-image-based rendering(DIBR) forvirtual view synthesis is the key technology of multi-view video applications, whichmakes use of video datas of reference point and the corresponding depth images tosynthesize the arbitrary viewpoint, and the quality of rendered view largely dependson the quality of depth image.Unlike video image, depth image is a grayscale that contains a large number ofhomogeneous area, more space redundancy and sharp boundary. Using the traditionalvideo coding methods to compress depth image can produce serious distortion in theborder, which affects the quality of the synthetic view. This paper focus on theseissues, on the basis of the existing H.264/AVC video coding method and theproperties of depth image, proposes an edge adaptive down/up-sampling compressionmethod and depth image post-processing method. The main research focus on thefollowing two aspects:Firstly, in order to solve the problem of synthetic view distortion due to the edgeerror in the depth map compression and reconstruction process, this paper proposes anovel joint depth/texture edge adaptive up-sampling scheme to compress depth map.The algorithm takes into account the structual similarity between the low resolution(LR) depth map and the high resolution (HR) depth map as well as the edge similaritybetween depth map and the corresponding color map. The first step is to detect thehorizontal and vertical edge of LR depth map and divide the whole depth map intothree regions. The second step ultilizes the information of color map to direct thereconstruction of accurate HR depth map according to the characteristic of differentareas. Experimental results suggest that the proposed method improves both depthmap coding efficiency and perceptual quality of rendered views.Secondly, aiming at the coding distortion problem in the process of depth map compression using the traditional video coding standard, this paper propose a post-processing method based on weighed mode filter (WMF). In the first step, identify theedge of depth map and color map by edge detection algorithm. Then get the real edge,the artificial edge caused by coding distortion and the smooth area through depthdiscontinuity analysis. And adaptively choose the filter according to the importance ofdifferent regions to eliminate the noise produced by the coding compression on thepremise of protecting the real edge. Experimental tests show that the proposed methodcan effectively improve the subjective and objective quality of the reconstructed depthmap and synthsized virtual view.