Solid model construction from three-dimensional images.

In many situations data describing an object are in the form of a 3D image, e.g., output from topology optimization methods or medical images. However, design and manufacturing tasks require a solid model as input; therefore an efficient, automatic method for converting a 3D image to a solid model is necessary. Previous methods have been limited to simple objects and have required significant user guidance. This dissertation presents a new method for constructing a solid model from a 3D image that overcomes these problems. In order to construct a solid model from a 3D image, five tasks must be performed: grouping, transition formation, edge extraction, parameter correspondence, and surface fitting. Grouping subdivides the voxels in the 3D image into branches and transition formation inserts intermediate image slices into the 3D image in order to close the gaps at the junctions between branches. Then edges are extracted from each image slice, forming contours. Two parameter correspondence schemes---boundary and shape based---are used to assign global parameter values to the contour points. Curves are fit through the contours and NURBS skinning surfaces are fit through the curves in a branch. The ends of the skinning surfaces are capped with planes and the branches are joined with the Boolean union operator to form a B-rep solid model. Finally, methods are proposed to capture information about material properties available in the 3D image within the solid model. New algorithms to perform the five tasks are presented. By using image processing techniques to perform these tasks, we are able to construct solid models of geometrically and topologically complex objects quickly and with very little user interaction. In addition, we discuss types of functions that can be fit to material data and present a new way to map a complex 3D region onto a cylinder in order to facilitate function fitting. The algorithms are used to construct solid models of medical artifacts and novel mechanisms and microstructures.PhDApplied SciencesComputer scienceHealth and Environmental SciencesMechanical engineeringMedical imagingUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/131720/2/9929892.pd