Stereo matching and belief propagation on singly connected graphs

Loopy Belief Propagation has revealed to be more than a very promising technique for solving early vision problems, such as stereo, by inference. However, cyclic graphical models are still a major challenge, since they introduce approximation issues which could affect the accuracy of the results. Breaking loops could be then the proper way to overcome those problems. In this paper, we will show how to transform an MRF grid into a suitable set of singly connected graphs. Then, a propagation schema, based on that set, is described. Experiments performed on different stereo pairs will prove the accuracy of our approach with respect to current state-of-art methods.