A fast segmentation of MRI image based on chan-vese model

Nowadays, active contour model and level set method have made a great success in the image segmentation, but these methods also have drawbacks: existence of local minima because of non-convexity and the huge amount of calculation. To solve these problems, the paper improved the original Chan-Vese model. Firstly, the nonconvex Chan-Vese model can be reformulated as convex optimization problem based on the work of Chan, et al. This can extract a global minmizer of the model. Then edge detector operator was incorporated into convex Chan-Vese model, and a hybrid model based on edge and region information was proposed. As for the drawback of level set method, the paper exploited fast duality projection algorithm to compute the global minimum of the proposed model. Meanwhile, the paper proposed a new iteration terminal condition which advoided the useless iterations and reduced the iteration time. Finally, the proposed model is applied to MRI images, and the result proves that the proposed model can extract the target region rapidly and accurately.