Tensor bidirectional reflectance distribution function model graph based multi-spectral face recognition

Multi-spectral face recognition can acquire good performance under illumination variation, but not for pose variation. As human faces are neither exactly Lambertian nor entirely convex and hence previous Lambertian assumption based multi-spectral face recognition fall short when dealing with pose variation. In this paper, a tensor bidirectional reflectance distribution function (BRDF) model graph based multi-spectral face recognition method is proposed to acquire good recognition performance under pose variation. First, face is divided into several feature regions according to the spectral characteristics. Then tensor spline is used to model the BRDF of every face feature region, which takes the pose, spectral and spatial information into consideration. Third, according to the relationship among these feature regions, a tensor spline BRDF relationship graph is constructed to model the characteristic of the face and used for face recognition. The trials of our experiment are conducted on multi-spectral face data-based, it is acquired using a CCD camera equipped LCTF (cover the spectral range from 400nm to 720, and be separated into 33 bands). And we compare proposed method with previous Lambertian assumption based multi-spectral face recognition method and 2DPCA face recognition method, and demonstrate experimentally that this algorithm can be used to recognize faces over time in the presence of changes in facial pose.