SciCADE 2013
International Conference on Scientific Computation and Differential Equations
September 16-20, 2013, Valladolid (Spain)

Invited Talk

Show full list of Invited Contributions to Minisymposia Show talk context (MS18)


A nonconvex model for image segmentation

B. Coll, J. Duran and C. Sbert

Abstract
In this paper, we study the problem of unsupervised segmentation through the minimization of energies composed of a quadratic data-fidelity term and a nonconvex regularization term. Since probably the greatest amount of information in an image is contained in the edges of the objects, we give some conditions for the design of an edge-preserving regularization. We show that, under those edge-preserving conditions, the functional can be rewritten in a dual formulation as a weighted total variation and thus, Chambolle's projection algorithm applies. This leads to design an efficient algorithm based on alternate minimizations on the two variables. Experimental results are presented and show the effectiveness and the efficiency of the proposed algorithm.

Bibliography
[1] B. Coll, J. Duran and C. Sbert, A Duality Based Approach for Nonconvex Image Segmentation, Preprint 2012.

Organized by         Universidad de Valladolid     IMUVA