Facial Shape-from-shading and Recognition Using Principal Geodesic Analysis and Robust Statistics

摘要: The aim in this paper is to use principal geodesic analysis model the statistical variations for sets of facial needle maps. We commence by showing how represent distribution surface normals using exponential map. Shape deformations are described on Using ideas from robust statistics we show deformable may be fitted images which there significant self-shadowing. Moreover, demonstrate that resulting shape-from-shading algorithm can used recover accurate shape and albedo real world images. In particular, effectively fill-in when more than 30% its area subject To investigate utility parameters delivered method, conduct experiments with illumination insensitive face recognition. present a novel recognition strategy similarity measured space parameters. also recovered information generate normalized prototype performed. Finally that, single input image, able basis employed number well known illumination-insensitive algorithms. geodesics provide an efficient parameterization harmonic