Résumé en anglais

A new approach of edge integration for shape modeling is presented. Detection of the boundaries of objects in a given gray--level image is performed by a multi--stage procedure that integrates a potential generator and a weighted distance transform. This enables us to find the global minimum of an active contour model's energy. This approach has an advantage over the classical snakes models since initialization becomes an easy task and the curve is not trapped at a local minimum. Applying the procedure enables the detection of object boundaries between two given points in cases of open boundaries like roads crossing the image domain, and closing a contour that models the boundary of an isolated object in the image, given only one point close to the objects' boundary. Closing contours is performed by using a topology--based saddle search routine. We show examples of our method applied to real images, typical for snakes applications.