Depth Map Coding And Depth-based View Rendering In Free-view Video System

FVV (Free-view video system) is an innovative visual media that enables us to view a 3D scene by freely changing our viewpoints. And now it is the hot point in video signal processing research. This thesis mostly makes research on depth map coding and DIBR(depth image based rendering) technology.Firstly, An view synthesis oriented depth coding algorithm is proposed. Considering the influence of the flicking problem in the backguround, we use the encoding mode of the color image, the object edge at the depth map and the frame difference of the two successive color frame to segment the map into different regions. Based on the stat that different region has the different coding modes in the depth map, macroblocks in different regions are encoded with different RDO schemes so as to speed up the encoding process while keep the quality of the reconstructed depth map. Experimental results show that the proposed scheme not only improves the encoding speed significantly, but also saves bitrate of depth map sequence while maintaining high quality of synthesized virtual view images.Secondly, in FVV system virtual view genetation can be accomplished by DIBR technique. Considering the fact that the quality of the synthesised virtual viewpoint images will be greatly affected by the boundary areas in the depth map. So we proposed a new scheme to protect the object edge at the foundation of the view synthesis oriented depth coding algorithm. Experimental results show that the proposed method not only improves the encoding speed significantly, but also increase the quality of synthesized virtual view images.Thirdly, two rendering schemes based on spatial correlation are proposed to speed up the rendering while maintain the quality of rendered virtual image. We present a new block-based view synthesis algorithm to reduce the mapping points in order to reduce the computing time. Next, an object based delaminated warping is proposed to improve the warping quality. For solve the blur at the object edge while using the Inpainting algorthim to fill the hole, we proposed a new hole filling scheme. The experimental results show that blocked-based rendering can well improve the rendering speed. The object-based delaminated warping and the new hole filling algorithm can improve the subject quality of the virtual view.Finally, one rendering scheme based on spatial and temporal correlation of the sequence was proposed. Based on the fact that points with the same depth at two successive frame keep the same position at the virtual view through 3D-warping. we use the depth coding modes to find out the static region at the two successive frames. Then, we just copy the mapping relation to reduce the mapping points, so as to improve the speed. In addition, we use the virtual depth to detect the mistake in the virtual color image, and the median filter is used at the mistake points. The experimental results show that the proposed algorithm not only improve the speed, but also improve the subject quality of the virtual view.