Stereo correspondence using stripe adjacency graph

This paper presents a new approach to solve the problem of stereo correspondence. In order to extract visible surfaces with similar texture from two different images, our method encodes input images to a series of so-called Stripe Adjacency Graphs (SAG). A stripe is a connected region with only one segment on each scan-line and a SAG is a group of stripes with neighbourhood relationships. The algorithm retrieves surfaces from binocular images according to the matching degree between two sub-graphs in two images. The extracted surfaces are evaluated by the global matching cost that is defined as the sum of inter-stripe and intra-stripe energy. The experimental results show that our algorithm is a fast as well as an effective algorithm. It can give a dense disparity map.