The Split Bregman algorithm of Chan-Vese model without re-initialization

In this paper, we designed the Split Bregman algorithm for Chan-Vese model (i.e., active contours without edges) without re-initialization. By introducing an auxiliary vector variable, a vector Bregman parameter, and using alternating minimization technique, original optimization problem of the classical variational image segmentation model is transformed into two sub-problems of minimization in an alternating form. The former is a simpler PDE that can be solved by a more conventional finite difference scheme; the latter is a generalized soft thresholding formula in analytical form. It means we can build a more efficient algorithm than the existing numerical algorithm built on the traditional PDE. We apply the proposed algorithm to both simulated and real images with different features and get promising results.