a generative appearance-based method for recognizing human faces under variation in lighting and viewpoint. The method exploits the fact that the set of images of an object in fixed pose, but under all possible illumination conditions, is a convex