A system which provides a total matching of a stereo pair of images is described: to every point in the first image a corresponding point in the second image is assigned. The processing consists of determining a limited number of matching candidates for each point in the first image and ordering the possible matches of a point for decreasing value of likelihood. Geometrical constraints are applied to determine the consistency among matches associated with different points. The assumption that points close in the image correspond to points close in space (continuity assumption) has been used. In most of the cases the scenes are composed of objects in contrast with the background.The target image is then preprocessed in order to extract the objects and then to reduce the areas to be matched to the object areas. The algorithm has been successfully applied to artificial scenes and real scenes. Examples and results are presented.