Level set technique, Image Segmentation and Cues Integration : Concepts and Recent Results.

Rachid Deriche (INRIA Sophia Antipolis,France)
This talk presents an overview of the level set based variational image segmentation models I have recently developed. The methodology extends the Classical and the Geodesic Active Contours concepts to provide a common framework to segment multivalued images while integrating different cues (color, texture, motion, shape priors..). The use of boundary as well as region information and the level set technique are the main distinctive features. This framework is currently extended to deal with Volumetric as well as non flat images i.e defined on a manifold. Various results will illustrate the potential of the approach.