Reconstructing curved surfaces from specular reflection patterns using spline surface fitting of normals

Abstract

Apple Computer, 1 Infinite Loop M/S 301-3J, Cupertino, CA 95014. Computer Science Division, University of California, Berkeley, CA 94720-1776. www.cs.berkeley.edu/projects/optical/ School of Optometry, University of California, Berkeley, CA 94720-2020. We present an algorithm that reconstructs a threedimensional surface model from an image. The image is generated by illuminating a specularly reflective surface with a pattern of light. We discuss the application of this algorithm to an important problem in biomedicine, namely the measurement of the human cornea, although the algorithm is also applicable elsewhere. The distinction between this reconstruction technique and more traditional techniques that use light patterns is that the image is formed by reflection. Therefore, the reconstruction algorithm fits a surface to a set of normals rather than to a set of positions. Furthermore, the normals do not have prescribed surface positions. We show that small surface details can be recovered more accurately using this approach. The results of the algorithm are used in an interactive visualization of the cornea.