Biometric technologies have a great potential because they identify subjects by their characteristics rather than passwords can be divulged or compromised. Face recognition has distinct advantages in terms of utility, since it can be done passively and unobtrusively at a comfortable distance. However, current 2D face recognition solutions lack sufficient robustness because 2D facial images are not invariant to pose and illumination conditions.

In this dissertation, a hybrid approach to identification/verification that employs 3D data for enrollment and 2D for verification/identification is presented. During the enrollment process, an annotated face model (AFM) is fit to the subject's 3D data. In the authentication process, the AFM is relit using a bidirectional relighting algorithm to the illumination in the input 2D image (assuming a known pose) and then it compares the data. We have tested our algorithm using a database containing 23 subjects and 1676 probes, on which we obtained 12.1% EER.