Title: Adaptive Surface Reconstruction for 3D CT-Data based on Geometric Modelling
Name: Dennis Mosbach
Phone: +49 631 31600-4859
Given a 3D volume image, obtained by computed tomography scanning, we are interested in obtaining a model for the scanned object. The standard approach would be to apply the marching cubes algorithm to create a triangulation of the material interfaces. In practice this method has several drawbacks. Especially for high resolution images and objects with a large surface area, the resulting triangulation may contain too many triangles to be handled efficiently in subsequent processing steps like analysis or simulations. This problem can be treated by mesh simplification algorithm or adaptations to marching cubes. However, in practice, these modifications can lead to accuracy losses in the results.
As a new approach, we investigate the use B-Spline surface patches to model the scanned objects. They require far less disc space and can be triangulated in any user desired resolution. Furthermore, geometric properties, like tangents or curvature, can easily be computed. The computation of B-spline surfaces from volume images is closely related to approximating point clouds with B-splines. Therefore, methods for point cloud approximation will serve as template. If additional geometric data is available (e.g. CAD models that were used to fabricate the scanned objects), it might also be worth to investigate whether it can be used to improve the quality of the results.
We expect to produce smooth analytical descriptions of the scanned objects. The resulting models should capture the shape of the object in general as well as fundamental characteristics like topology, sharp features, and curvature behavior. Also, if desired by the user, it will be possible to compensate small errors, like noise, in the input data.
Marching cubes (left) and B-spline model (right) of an asymmetric torus.