Local information based level set methods are comparatively effective in segmention of images with intensity inhomogeneity which often occurs in real images. However, there are problems such as local minima in the active contour energy and the considerable computing-consuming in these models in practice. A novel level set method based on improved region-scalable fitting is proposed for such image segmentation. The global information containing local feature term, measured by Bregman divergence, is utilized to accelerate the contour evolution when the contour is far away from object boundaries, in order to improve the robustness of the algorithm to the initial position. The local feature item of the RSF model is used to promote the convergence of the contour curve to the boundary of the object, thereby increase the capability to cope with intensity inhomogeneity. Finally, comparative experiments on synthetic images, mediation images and the real images validate the performance of the algorithm.