Genetic Algorithm to Set Active Contour Jean-Jacques Rousselle, Nicole Vincent, Nicolas Verbeke Laboratoire d'Informatique (LI) Université François Rabelais de Tours 64, av. Jean Portalis, 37200 Tours, France (rousselle, vincent)@univ-tours.fr Abstract. Active contours, very popular in image segmentation, suffer from delicate adjustments of many parameters. We propose to carry out these adjustments using genetic algorithm. Here an active contour is implemented using a greedy algorithm. Within this framework, two approaches are presented. A supervised approach which delivers a global set of parameters. In this case the greedy algorithm is involved in the evaluation function of the genetic algorithm. The second approach is unsupervised. It determines a local set of parameters. The genetic algorithm computes a set of parameters which minimizes the energy at each point in the neighborhood of the current point in the greedy algorithm try to move. 1 Introduction In image processing, segmentation is an essential phase. The traditional models detect some points of contour which are needed to be chained. This chaining is no more needed when using an active contour "snake". The snake is a curve, deformable under the influence of various forces. These forces must be balanced by the user according to the object of interest and to the images to be analyzed. Generally, the adjustment of these parameters is carried out by trial and error. It is recognized that these adjustments are difficult, but few authors were interested in that problem. We already proposed an approach by experimental design [6] and a method using random parameters search [7]. Here we will use a genetic optimization method. It is based on the natural selection theory. After having presented active contours then the genetic algorithm technique, we will explain the results of two approaches: one supervised and the other unsupervised. 2 Active Contour The concept of active contour, or snake, comes first from Kass et al.'s works [4]. It is a question of evolving a set of ordered points (describing a closed contour) so that it is nearest as possible to the contour sought for in the image. Such a set of points is called "snake" and the points which make it up are "snaxels". Let V be a snake that is composed of n points and defined as follows: 10th International Conference Computer Analysis of Images and Patterns CAIP'2003, 25-27 août 2003, Groningen, Hollande.