Research On 3D-HEVC Depth Map Preprocessing And Error Concealment Techniques

Since multi-view video plus depth(MVD) has good standard compatibility, and can be utilized to synthesize virtual views to enable stereoscopic free view-point television(FTV) in the decoder, it has been chosen to be the standard data format of the emerging three dimensional video(3DV) coding standard by moving picture experts group(MPEG). 3DV coding standard can be developed based on either H.264/AVC or H.265/HEVC standard. Compared with H.264/AVC, the HEVC, as the new generation standard made by ITU-T and ISO jointly, can save almost half coding bit with the same constructed video quality, therefore, 3D-HEVC arouse more concern. The involving of depth maps increases the coding bit-rate, how to increase the depth coding efficiency becomes the key of 3D-HEVC. Besides, with the increase of coding bit-rate, the risk of network congestion increases in the limited network, causing the loss of video data and the error in decoding. Therefore, it is essential to make more efficient error concealment technique for 3DV. This paper focuses on the preprocessing of depth map to improve efficiency of coding and 3DV error concealment. The main content of this paper is as follow:With the limit in depth estimation algorithm, the obtained quality of depth map is low. It shows in several aspects. The consistency of object edge information between depth map and video picture is bad, causing many false edges. False edges not only increase coding bitrate, but decrease the quality of synthetic view in the decoding end. To remove the false edges and improve the coding efficiency of depth map, a preprocessing of depth map is proposed. By using the correlation of object edges between depth map and video picture, false edges are smoothed with anisotropic diffusion method. Due to the characteristic of anisotropy in proposed method, the real edge information with high correlation of object edges between depth map and video picture can be protected effectively. On the premise of keeping the quality of synthetic view, this method can efficiently remove false edges and improve depth map coding efficiency.For the lost frames information in the network transferring, the temporal correlation of frames before and after is used to implement error concealment, but fails to recover correctly for the moving foreground. Compared with 2D video, the interview correlation of MVC information is also used to interpolation of lost frames. An interpolation technique to differentiate foreground and background is proposed to realize 3DV error concealment. Firstly, the difference between frames before and after is compared with threshold to differentiate foreground and background, and decreased noise by morphological filtering. Interpolation for frames before and after is still utilized for background. While the moving foreground uses the synthetic view of neighbor in current view to fill. The proposed method comprehensively utilizes temporal and interview correlation, and improve the quality of reconstructed frames.