This paper deals with the problem of automatic reconstruction of contour lines out of scanned maps. Our approach is essentially based on the global topology of the contour lines (ie. a set of non-intersecting closed lines), but we use also some local and geometrical informations to guide the reasoning.
Our proposed method starts from a map scanned in RGB. The image is fisrt segmented by a colour classification process, and the objects are skeletonized end vectorized. Then, by a local analysis, we weight and add some hypothetical edges. By a relaxation process, we try to orientate the elevation curves, and so we could re-ajust the weight. Finally, we found a topologycally correct solution by resolving the dual problem (ie. a bicolouring problem).