Uncooled Infrared Imaging Face Recognition Using Kernel-Based Feature Vector Selection

This study considers an approximation to the Generalized Discriminant Analysis (GDA) and its applications to an uncooled infrared image face recognition problem. We consider the feature vector selection approach recently proposed by Baudat and Anouar, and combine it with the Linear Discriminant Analysis method (FVS-LDA). The resulting scheme is applied to the fifty-subject uncooled IR face database developed locally in an earlier project for comparison purposes. Identification and verification experiments are reported and compared to those obtained with the GDA implementation. Results indicate that similar recognition performances may be obtained when using well- tuned FVS parameters for a significantly reduced computational effort.