Obtaining Intrinsic Shape Properties with a Disparity-Based Photometric Stereo

Abstract

Disparity-Based Photometric Stereo (DBPS) is a recently introduced process (Torreao and Pimentel, 1992) in which two or more photometric stereo images are matched to yield a photometric-disparity field which is akin to the disparity map generated in Stereoscopy. In (Torreão and Pimentel, 1992), it has been shown that, under appropriate conditions, the photometric disparities thus obtained can be employed in the recovery of a relative-depth map for the imaged surfaces. It is generally agreed, though, that for use as a shape representation framework, with the purposes of storage and recognition, a depth map is not the most convenient option, since it lacks the desirable invariance and generalization properties (Horn, 1986). (Even for biological vision systems, it hasn’t been firmly established that a depth map of surfaces is actually computed (Weinshall, 1991)).