Face Recognition Using 3-D Models: Pose and Illumination

摘要: Unconstrained illumination and pose variation lead to significant in the photographs of faces constitute a major hurdle preventing widespread use face recognition systems. The challenge is generalize from limited number images an individual broad range conditions. Recently, advances modeling effects have been accomplished using three-dimensional (3-D) shape information coupled with reflectance models. Notable developments understanding include nonexistence invariants, characterization set objects fixed under variable (the cone), introduction spherical harmonics low-dimensional linear subspaces for illumination. To novel conditions, either multiple must be available reconstruct 3-D or, if only single image accessible, prior about appearance general used. Morphable Model was introduced as generative model predict appearances while statistical on texture allowing its parameters estimated image. Based these new understandings, algorithms developed address joint challenges lighting. In this paper, we review provide brief survey resulting their performance