Experimental Evaluation of Eye Location Accuracies and Time-Lapse Effects on Face Recognition Systems

Abstract

It is claimed that eye location accuracy is very important to face recognition system performance. In most systems, the eye locations are the most significant facial landmark for the preprocessing step. Eye location estimates can be assessed in absolute terms (e.g., proximity to known eye location) and also in application-specific terms (e.g., performance of a system that employs the location). This paper assesses an automatic commercial eye-finding system in absolute and application-specific terms, using four different face recognition systems and a database of thousands of images. A pilot study on the time-lapse effect suggests that with the time-lapse increasing, the face recognition performance will degrade. Our experiments examine this effect by using a large image dataset, which has a time-lapse up to two years, with 250 subjects and 64300 probes. Experiment results show eye location accuracy is significant to face recognition system performance. Different systems can have different level of sensitivity, and the system using local feature analysis is less sensitive to eye location accuracy. Also all the algorithms tested in this study show that time-dependency exists in face recognition system.