Supervised Continuous Max-flow for Segmentation of Images with Intensity Inhomogeneities

Abstract

In this thesis we propose a regional adaptive active contour model for segmenting images with intensity inhomogeneities effectively. This model includes a regularization term, a global energy term and a local energy term. The regularization term is used to control the length of the border of the characteristic function. At the early stage, the global term which is larger than the local term can give us a good initial segmentation of the image quickly. In the later phase of the process, the local energy term dominating the whole function energy can localize the precise position of the object of interest in image with intensity inhomogeneities. An algorithm based on supervised continuous max-flow method is developed to solve this problem robustly and validly. A large number of experiments are presented to show how this proposed method behaves on different images.