Continuous Max-Flow for Image Segmentation with Shape Priors

Abstract

In this thesis we propose a stable method for image segmentation with shape priors. The original Chan-Vese intensity based segmentation model with regularisation term is extended to include shape prior information. We study shape priors which are pose invariant under the group of similarity transformations, that is under rotation, scaling and translation. In order to solve this problem robustly and effectively, an algorithm based on the theory of max-flow and min-cut is used in addition to a gradient descent procedure for updating the pose parameters. Comprehensive experiments are provided to demonstrate the behaviour of the proposed method on different images.