A Faster-converging Algorithm for Image Segmentation with a Modified Chan-Vese Model

We propose an algorithm for segmentation of grayscale images. Our algorithm computes a solution to the convex, unconstrained minimization problem proposed by T. Chan, S. Esedoḡlu, and M. Nikolova in [1], which is closely related to the Chan-Vese level set algorithm for the Mumford-Shah segmentation model. Up to now this problem has been solved with a gradient descent method. Our approach is a quasi-Newton method based on the lagged diffusivity algorithm [2] for minimizing the total-variation functional for image denoising [3]. Our results show that our algorithm requires a much smaller number of iterations and less time to converge than gradient descent, and is able to segment noisy images correctly. Keywords—image segmentation, Mumford-Shah segmentation, Chan-Vese model, quasi-Newton method