A new singular perturbation approach for image segmentation tracking

Deformable model based segmentation usually relies on a force field extracted from the image data through the computation of image gradient or gradient vector flow. At convergence, the work of the forces at the interface location should annihilate. This condition is not met in classical deformable formulations. In order to insure this condition, we previously introduced a constrained problem and a nonlinear approach in the framework of a deformable elastic template. From a computational point of view, theses two approaches can be very time consuming. Therefore, we propose in this paper a new simpler formulation using a singular perturbation technique. The nice behavior of the proposed model is demonstrated in the context of the segmentation and tracking of the heart contours in 2D cardiac MRI sequences.