Efficient view synthesis for multi-view video plus depth

Abstract

Depth-image based rendering is a popular technique for rendering virtual views for the multi-view video plus depth (MVD) 3DTV format. Instead of transmitting a high number of views to the receiver, only a few views are transmitted along with their corresponding depth maps and the missing views are interpolated from these transmitted views. This process is called view synthesis and has been the focus of a lot of research in recent times. In addition to the obvious quality constraints of view synthesis, focus needs to be laid on the efficiency and speed of the view synthesis process. In this paper, we propose a simple, highly efficient view synthesis process that utilizes a separable look-up table based technique to speed up the 3D warping process. We also develop a reliability-based weighting scheme for blending the candidate views into the final synthesized view. Simulation results show that the proposed method is approximately 6 times faster than VSRS 3.5 and has comparable objective quality. The proposed method is inherently parallel thus making it suitable to act as a framework for implementing high-speed view synthesis on a GP-GPU.