Segmenting surfaces of arbitrary topology: A two-step approach

We propose a two-step approach to segment closed surfaces in 3D of arbitrary topology. First, a pre segmentation step with an active contour method is performed. This evolution process does not take into account topology adaptions. Topologically correct segmentations are derived with Kazhdan's algorithm in a second step. Kazhdan's algorithm requires information on the surface normals, which are obtained from the active contour method. We show that the two-step algorithm is computationally efficient. Moreover, we apply the algorithms for segmentation of 3D ultrasound data.