Semi-implicit co-volume method in 3D image segmentation

We introduce a three-dimensional (3D) semi-implicit complementary volume numerical scheme for solving the level set formulation of (Riemannian) mean curvature flow problem. We apply the scheme to segmentation of objects (with interrupted edges) in 3D images. The method is unconditionally stable and efficient regarding computational times. The study of its experimental order of convergence on 3D examples shows its second order accuracy for smooth solutions and first order accuracy for highly singular solutions with vanishing gradients as arising in image segmentation.