Wide baseline stereo matching by corner-edge-regions

Abstract

Most matching methods assume that all pixels of a correlation window have equal depths and find correspondences by comparing this local window of pixels. However, this assumption is not valid at depth discontinuities and wide baseline images. In this paper, a new method to construct the local planar patches called corner-edge-regions (CERs) is introduced. Based on this adaptive plane identification method, a novel similarity measure for estimating tentative correspondence is proposed. Using this measure makes the matching system robust to severe occlusions and shrinking. The proposed matching method is tested on several wide baseline image pairs. Experimental results show our method can find more correct correspondences than other correlation window based matching methods. © Springer-Verlag 2004.