New insights into improving compression efficiency for distributed video coding

In this paper, we investigate a fundamental issue in the distributed video coding (DVC) that, once resolved, would substantially improve the compression efficiency in DVC. This fundamental issue is the underlining relation between the distribution of the prediction errors and the compression efficiency of DVC. In the current approach to DVC, after the construction of prediction frame at decoder, the difference between the prediction and the current frames, or the prediction error, is inversely proportional to the correlation between these frames. Most existing approaches in DVC attempt to maximize such correlation, or to minimize the prediction error, in order to achieve the desired video compression efficiency. Recently, the research in DVC has reached a plateau in terms of its performance in coding efficiency. We believe one key solution to take DVC out of such performance plateau will be to design a better scheme to represent the correlation more effectively in DVC. In this research, we have worked out compelling theoretical analysis and proved that, in order to reduce the number of bits to be sent to decoder, the distribution of the prediction errors needs to be as concentrated as possible. From the practical point of view, we also show that the error control codes adopted for DVC will achieve higher efficiency when the distribution of prediction errors is more concentrated.