A Fast Anisotropic Mumford-Shah Functional Based Segmentation

Digital (binary) image segmentation is a critical step in most image processing protocols, especially in medical imaging where accurate and fast segmentation and classification are a challenging issue. In this paper we present a fast relaxation algorithm to minimize an anistropic Mumford-Shah energy functional for piecewise constant approximation of corrupted data. The algorithm is tested with synthetic phantoms and some CT images of the abdomen. Our results are finally compared with manual segmentations in order to validate the proposed model.