8 - Détection de contours réguliers dans des images bruitées et texturées : une approche par contours actifs multiéchelle

This article deals with the detection of edges in noisy and textured images . After an overview of the shortcomings of local operators, we assess active contour models, which constrain the detection using a priori information . The implementation of active contours is difficult due to several problems . We suggest two methods which make this global approach more robust and easier to use . On one hand, the parameters are set using a calibration algorithm which relies on geometrical a priori . The value of the parameters depends upon the maximum curvature . On the other hand, a multiscale strategy reduces the need for an accurate initialization of active contours . The convergence is robust at coarse scales and the localization of edges is enhanced at fine scales . Several efficient algorithms are proposed to track contours over decreasing scale, using a prediction on the distorsion of boundaries caused by gaussian blurring . The experiments show the validity of our approach and the relevance of a cooperation between segmentation processes, especially to solve the initialization problem .