| Abstract | Most photometric stereo algorithms employ a Lambertian reflectance model witha varying albedo field and involve the appearances of only one object. The recovered albedos
and surface normals are object-specific and appearances not belonging to the object cannot
be easily handled. We generalize photometric stereo algorithms to handle all appearances
of all objects in a class, in particular the human face class, by assuming that albedos and
surface normals of all objects in the class be rank-constrained, i.e. lie in a subspace. Rank
constraints lead us to a factorization of an observation matrix that consists of exemplar images
of different objects under different illuminations. To fully recover the subspace bases or class-
specific albedos and surface normals, we employ integrability and face symmetry constraints
and propose a linearized algorithm. This algorithm takes into account the effects of the varying
albedo field by approximating the integrability terms using only the surface normals. We then
apply our generalized photometric stereo algorithm for recognizing faces under illumination
variations. As far as recognition is concerned, we can utilize a bootstrap set which is just a
collection of 2D image observations to avoid an explicit requirement that 3D information be
available. We obtain good recognition results using the PIE database.
|
