Multiclass Segmentation by Iterated ROF Thresholding

Variational models as the Mumford-Shah model and the active contour model have many applications in image segmentation. In this paper, we propose a new multiclass segmentation model by combining the Rudin-Osher-Fatemi model with an iterative thresholding procedure. We show that our new model for two classes is indeed equivalent to the Chan-Vese model but with an adapted regularization parameter which allows to segment classes with similar gray values. We propose an efficient algorithm and discuss its convergence under certain conditions. Experiments on cartoon, texture and medical images demonstrate that our algorithm is not only fast but provides very good segmentation results in comparison with other state-of-the-art segmentation models in particular for images containing classes of similar gray values.