An Examination and Comparison of Fairness of Face and Ocular Recognition Across Gender at NIR Spectrum

Published research suggests bias of face and ocular biometrics across demographics. Specifically, unequal performance is obtained for women and dark-skinned people. However, these published studies have examined the demographic bias in the visible spectrum (VIS). Factors such as makeup, facial hair, skin color, and illumination variation have been attributed to the bias of these systems at VIS. The near-infrared (NIR) spectrum offers an advantage over VIS in terms of robustness to factors such as illumination changes, makeup, and skin color. This chapter investigates and compares the bias of facial and ocular recognition systems at the NIR over VIS. Popular NIR-VIS datasets such as CASIA-Face-Africa, Notredame-NIVL, and PolyUIris are used to investigate the bias of facial and ocular biometrics across gender and spectrum. Grad-CAM-based explainable AI (XAI) technique is used to understand distinct image regions used for classification across modality and spectrum. This study provides further insight into the disparities in performance across gender and the potential of NIR technology in addressing the bias of biometric technology.