Similarities in African Ethnic Faces From the Biometric Recognition Viewpoint

Ogechukwu Iloanusi, Patrick J. Flynn, Patrick Tinsley; Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops, 2022, pp. 419-428

Face pose, illumination, and facial expressions are known factors that affect face recognition performance and have been studied at length in the literature. The impacts of demographic factors such as gender, race, and age on performance have also been studied, with increasing interest recently in the context of algorithmic bias concerns. This work is a study of face recognition performance using a database of faces of Nigerian subjects drawn from 28 different ethnicities. There are documented differences in facial anthropometric characteristics between some Nigerian ethnicities, and this study was intended to establish initial results regarding the impact of these inter-ethnic differences on face recognition performance. A comparison to performance on a database of Caucasian/Asian face images is made. Our study analyses how 28 African ethnicities affect face identification performance metrics by focusing on the genuine and impostor scores' distributions. Our analysis shows that face identification performance is not remarkably influenced by varying ethnicities within the African race though there are significant differences in relation to the Caucasian/Asian set.