Abstract.
In the present work, we consider regions matching of a pair of stereoscopic images. The originality of the proposed method lies in the use of the relative position of regions as a constraint instead of the conventional one based on the adjacency. Then, under the hypothesis that corresponding points in the two images lie on the same horizontal scanline, there exist at least two corresponding points which belong to two corresponding regions. Therfore, corresponding regions must overlap in y-dimension. To guarantee a great similarity between regions, we propose a new cost function which takes into account the local property at the vicinity of region boundaries. Thus, the matching process is carried out in two main steps. In the first step, the candidate matches which meet the relative position and the overlapping constraints are extracted. In the second step , the correct matches which are solution of the minimization of the proposed cost function are searched. To illustrate the performance of the proposed algorithm, experimental results on synthetized and real stereo images are given.

Keywords: Stereo vision, Regions matching, Segmentation, Relative position constraint, Overlapping constraint.