Geometrically Deformable Models for Automatic Model-based Segmentation

In this paper we adress the problem of extracting geometric objekts from 2D and 3D datasets, given a reference shape pr model of the object. Today automatic segmentation can be a problematic due to noise and artefacts which are inherent in most images. The segmentation problem in these images cannot be solved adequately without the use of knowledge. In contrast to many other approaches which model a priori information via their parameterization, we propose its modeling in an energy minimizing context. Therefore, we extend the concept of geometrically deformable models by integration of given a priori knowldge. The proposed method increases the robustness of segmentation by processing this additional information during the whole segmentation process and not only at the initialization. We have applied this method to the reconstruction of teeth from range data.