Abstract

Modern image data collection techniques such as computed tomography, scanning electron microscopy, and digital stereoscopy make possible the computation of three-dimensional (3D) structure of scenes or objects. Such images are encountered in medical imaging, modeling of physical phenomena, geometric modeling and computer vision. There is a growing need in all these fields for efficient representation schemes of the 3D digital image. Of special interest in image description and modeling are surface methods, of which there are two general types. One is an interpolative method which expresses the surface by subdivisions known as surface primitives or patches ^(1,2). It is used in computer aided geometric design and computer graphics. The other is concerned with describing surfaces "in the large", mainly in terms of shape descriptors or globaly defined functions of the boundary⁽³⁾. In either case parametric techniques are preferred since the representations are of lower dimensionality, are axes independent, and unambiguous for multivalued surfaces. Unfortunately 3D spatial data is most often encountered as objects in cellular space and the problem of parametrizing an arbitrary surface cannot in general be solved^(4).

© 1987 Optical Society of America