Advanced residual prediction enhancement for 3D-HEVC

Advanced residual prediction (ARP) is an efficient coding tool in 3D extension of HEVC (3D-HEVC) by exploiting the residual correlation between views. In the early version of ARP, when current prediction unit (PU) has a temporal reference picture, a reference block is firstly identified by a disparity vector and the residual predictor is then produced by aligning the motion information associated with current PU at the current view for motion compensation in the reference view. However, ARP is not allowed when current PU is predicted from an inter-view reference picture. Furthermore, the motion alignment during the ARP is done in a PU level, thus may not be good enough. In this paper, an enhanced ARP scheme is proposed to first extend ARP to the prediction of inter-view residual and then extending the motion alignment to a block level (which can be lower than PU). The proposed method has been partially adopted by 3D-HEVC. Experimental results demonstrate that the proposed scheme achieves 1.3 ~ 4.2% BD rate reduction for non-base views when compared to 3D-HEVC anchor with ARP enabled.