An original algorithm is presented that matches the segments of a pair of stereoscopic images and then improves the edge segmentation of images to make the interpretation process easier. The method does not consider each segment independently to perform the matching, rather it groups and matches the segments simultaneously. First, the linked segments are grouped and matched. Then, the same process is repeated while considering the segments more and more distant. At each step, different criteria such as segment connection type and relative position of the segments are used to select the correct matchings among ail the matching hypotheses. The segmentation improvement takes into account the geometric and topologic properties of objects-the computed sets of segments are considered the projection of the faces of these objects. The segmentation improvement method relies on the comparison of the matched sets obtained previously.