Active contours driven by non-local Gaussian distribution fitting energy for image segmentation

Abstract

Image segmentation is still a challenging task in image processing field because of unpredictable noise and intensity inhomogeneity in images. In this paper, we present a novel active contour model for image segmentation by constructing a robust truncated kernel function. It utilizes image patches to perceive the neighborhood intensities of pixel at the same time considers the spatial distance within a local window. By using this truncated kernel function, the proposed method can accurately segment images with intensity inhomogeneity while guaranteeing certain noise robustness. Extensive evaluations on synthetic and real images are provided to demonstrate the superiority of our method. The model makes full use of image patch information to strengthen the robustness against noise and intensity inhomogeneity in images.