Compressive sensing based multiview image coding with belief propagation

Multiview image acquisition systems usually involve many closely-located cameras. In some scenarios, it might be possible to significantly reduce the sample rates of some cameras and still reconstruct the corresponding images with good quality, by taking advantage of the side information from neighboring views. In this paper, we investigate the application of the belief propagation-based compressive sensing (CS-BP) theory to these cameras. However, the original CS-BP algorithm assumes that all unknown variables have the same prior distribution, which is not true in many cases, especially images. We show in this paper how to generalize the decoding of the CS-BP method such that it can fully utilize the side information and handle variables with different distributions. Preliminary numerical results with both 1-D and 2-D data demonstrate that the proposed generalizations can significantly improve the performance of the CS-BP method.