Characterization of Human Faces under Illumination Variations Using Rank, Integrability, and Symmetry Constraints

摘要: Photometric stereo algorithms use a Lambertian reflectance model with varying albedo field and involve the appearances of only one object. This paper extends photometric to handle all objects in class, particular class human faces. Similarity among facial motivates rank constraint on albedos surface normals class. leads factorization an observation matrix that consists exemplar images different under illuminations, which is beyond what can be analyzed using bilinear analysis. Bilinear analysis requires same illuminations. To fully recover class-specific normals, integrability face symmetry constraints are employed. The proposed linear algorithm takes into account effects by approximating terms normals. As application, recognition illumination variation presented. enables separate source from observed appearance keep illuminant-invariant information appropriate for recognition. Good results have been obtained PIE dataset.