Depth map recovery for multi-view using belief propagation

Depth maps are playing an important role in the multi-view system. For multi-view, depth maps can be used not only for image-based rendering (IBR), but also for multi-view stereo reconstruction by merging depth maps into 3D models. In this paper, we propose a novel approach to recover depth maps for multi-view. The most contribution of our work is the combination of a robust photo-consistency metric and the belief propagation algorithm, which results in several advantages. First, with the robust photo-consistency metric, our approach effectively utilizes the depth information from the multiple views and solves the occlusion problem as well. Second, the depth values of pixels in textureless regions can be effectively computed by belief propagation with depth discontinuity concerned. Finally, the proposed approach can deal with various scenes. The experimental results show that our approach can recover satisfactory depth maps.