F. Abram, P. T. Sander

Thème Images/Réalité Virtuelle
Laboratoire UNSA/CNRS I3S
Les Algorithmes, Bâtiment Euclide
2000, route des Lucioles
Sophia Antipolis 06410 BIOT France


In many cases, the general overall shape of the contours of specific objects in images is known, e.g., bones and organs in biomedical images. This information can be used to initialize an active contour and to constrain its global evolution while fitting itself by local deformations to the object in the image. The model thus not only extracts the object in the image, but also provides an evaluation of the deformation of the extracted contour with respect to the given original shape. This paper describes an active contour segmentation technique based on such knowledge of contours in images. Following a rough initial placement of the model, the deformation process consists in three stages, a free deformation stage under the effect of image attraction or other external forces, a stage of regularization to constrain the previous deformation with respect to the a priori known shape of the contour, and finally a stage of re-positioning the contour using the theory of material systems.

Keywords: active contours, edge detection, segmentation, registration, regularization, material systems, free form deformation, differential geometry.