On the basis of the Chan-Vese model, a new splitting active contour method for image segmentation is presented. The main idea following is to divide an image into two parts at every iteration, which is similar to the procedure of cell splitting. Then, the model is able to detect all the objects or details in the image. In addition, it enjoys the merit of processing any specific region in the image, even the inconsecutive one. This directly leads to the improvement of computing efficiency whereas segmentation is limited to region of interest (ROI) rather than the whole image. Furthermore, due to the regional constraint of operation, our model outperforms the existing multiphase Chan-Vese model in terms of sensitivity to the initialization. The principle of our model is described in detail, and the method is implemented under the level set framework. Experiments on both synthetic and medical images are carried out, and the comparative results to Chan-Vese model and multiphase Chan-Vese model are also shown.