Eigenface-based facial recognition systems rely heavily on predetermined eye locations to properly orient the input face prior to template generation. Gross errors in the eye detection process can be identified by examining the reconstruction image of the resulting eigenspace representation. Subtle variation in the precision of eye finding that does not prevent subsequent enrollment has not been effectively studied or reported by the biometrics testing community. We quantify the impact of eye locations on face recognition match scores for identical subject images. The scores are analyzed to better understand the consequences and sensitivity of eye finding for more general applications when eye locations must be determined automatically.

Download Full Presentation