Rate distorsion analysis in a disparity compensated scheme

This paper addresses the problem of rate distortion analysis in the context of multi-view image coding, where images are predicted via disparity compensation based on depth map. We first present an analytical model for the variance of the residual error in a predicted frame when the prediction is done with the help of a compressed depth map. This residual variance model presents a convenient expression that separates the different error origins (reference frame quantization, depth map coding, and motion activity). We then validate the novel analytical model by testing separately its different underlying hypotheses. Finally, we illustrate an application of our analytical model in a simple bit allocation problem where the objective is to determine the optimal distribution of a global bit budget among reference frame, depth map and disparity-compensated frame. We observe that the optimal allocation given by the analytical model corresponds in practice to the best rate distribution for high bitrate, which confirms the potential of the proposed model in the design of rate-controlled multi-view coding algorithms.