Dealing with segmentation errors in region-based stereo matching

Graph-based matching methods have been widely used in the areas of object recognition and stereo correspondence. In this paper, an algorithm to deal with segmentation errors in region-based matching is proposed, which consists of a preprocessing stage to the classical graph-based matching algorithm. Some regions are merged and included in the matching process in order to avoid the di!erences in segmentation. The selection of an appropriate similarity criterion to create the initial nodes in the graph and the use of approximative algorithms to "nd maximal cliques are important issues in order to reduce the computational burden. The experimental results show that the method is robust enough in the presence of noise. ( 2000 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.