Simplified reference pixel selection for constant partition value coding in 3D-HEVC

To accurately represent object edges in depth maps, new depth intra prediction modes, namely Depth Modelling Modes (DMMs), are introduced into 3D-HEVC. In the DMMs, a depth block is approximated by a model that partitions the block into two non-rectangular regions, where each region is represented by a constant value referred to as Constant Partition Value (CPV). To code the CPV, a predicted CPV for each region is obtained by using the adjacent pixels of available neighboring blocks. However, the current reference pixel selection method may fail to find a good prediction in some cases, as to be shown in this paper. Accordingly we propose a simplified yet more effective reference pixel selection method to streamline the CPV prediction. While simplifying the reference selection process in terms of computational complexity, the proposed method also demonstrates 0.07% and 0.06% BD-rate saving for synthesized views, under all-intra case and random-access case, respectively.