Allowable depth distortion based depth filtering for 3D high efficiency video coding

Depth videos shall be efficiently compressed and transmitted to the client for view synthesis in Three-Dimensional (3D) video system. Since depth video may contain noise that reduce the coding efficiency, we propose a depth filtering algorithm for 3D depth coding, which exploits the Allowable Depth Distortion (ADD) in view synthesis and is able to improve the coding performance of the depth encoder. Firstly, the depth values has the same rendering position based on the ADD model are clustered. Then, the clustered depth are filtered and set to the optimal depth value for each group by minimizing the view synthesis error. The filtered depth videos are smoother and can be more effectively compressed by the existing 3D High Efficiency Video Coding (HEVC) depth encoder. Experimental results show that the proposed depth filtering method can assist the depth encoder achieve 5.87% bit rate reduction in terms of Bjonteggard Delta Bit Rate (BDBR) and 0.25dB quality gain in terms of Bjonteggard Delta Peak-Signal-to-Noise Ratio (BDPSNR) on average as compared with that of coding the original depth maps.