Searching correspondences in colour stereo images-an approach using the fuzzy integral

We introduce an approach for obtaining dense contour maps of objects, which is based on the combination of the images of two or more colour cameras. As with all stereo techniques corresponding points in the image planes of the cameras, i.e. pixels illuminated by a certain point on the object, must be identified before the distance between the cameras and the object point is computed by triangulation. The search process should find as many correspondences as possible to enable the generation of dense range maps and the process must be accurate because errors in the pixel assignment directly affect the precision of the computed distance. With the correspondence search presented here the function that determines the similarity between the image segments under consideration is a special form of the fuzzy integral and hence nonlinear. Through fuzzy measures knowledge about the colour channels, e.g. their transfer function or their credibility, may be incorporated into the search process. Our first findings, which are discussed in the paper, indicate that the method is considerably more effective than the standard approach of minimizing the squared difference between line blocks. We conclude the paper with a discussion of the potential of the method and directions of possible future research