Using One Graph-Cut to Fuse Multiple Candidate Maps in Depth Estimation.

Abstract

Graph-cut techniques for depth and disparity estimations are known to be powerful but also slow. We propose a graph-cut framework that is able to estimate depth maps from a set of candidate values. By employing a restricted set of candidates for each pixel, rough depth maps can be effectively refined to be accurate, smooth and continuous. The contribution of this work is to extend the graph structure proposed in the original papers on Graph-Cuts by Ishikawa and Roy, in such a way that sparse sets of candidates can be handled in one graph-cut.