Level Set Based Multispectral Segmentation with Corners

In this paper we propose an active contour model for segmentation based on the Chan-Vese model. The new model can capture inherent sharp features, i.e., the sharp corners of objects, which are often smoothed by the regularization term in segmentation. Motivated by the snaked based method in [M. Droske and A. Bertozzi, SIAM J. Imaging Sci., 3 (2010), pp. 21-51] that emphasizes straight edges and corners without regard to orientation, we develop a region based method with a level set representation. The model combines the Chan-Vese model with the level set version of a higher order nonlinear term. We extend this model to multispectral images. Higher order methods can be very stiff, so we propose a splitting scheme to remove the stiffness and prove the model's stability and convergence. Finally we show numerical results on gray, color, and hyperspectral images. We can see that the model is robust to noise.