Representation and Description of Complex 3D Objects from Medical Volume Data

The representation and description of 3D objects from volume data is a quite new field in computer vision. In the past years the analysis of 3D medical data has been largely limited to different low level segmentation and 3D visualization techniques. But there is more and more need for a change of representation which results in a symbolic description of complex 3D objects, giving access to qualitative and quantitative parameters. This paper presents an algorithm which generates automatically a decomposition and a symbolic description of 3D objects from volume data. Complex objects are decomposed using morphological shrinking and reexpansion operations. The resulting set of compact subobjects is approximated in terms of volumetric primitives and sets of elliptical cylinders. The output description is unique for a certain object in different spatial position and orientation. In this sense, the generated description is invariant of translation and rotation of the object. The algorithm has been successfully applied to image data of the bone structure of a human knee which was acquired in medicine by computed. tomography (CT) imaging.