Face Quality Estimation and Its Correlation to Demographic and Non-Demographic Bias in Face Recognition

Face quality assessment aims at estimating the utility of a face image for\nthe purpose of recognition. It is a key factor to achieve high face recognition\nperformances. Currently, the high performance of these face recognition systems\ncome with the cost of a strong bias against demographic and non-demographic\nsub-groups. Recent work has shown that face quality assessment algorithms\nshould adapt to the deployed face recognition system, in order to achieve\nhighly accurate and robust quality estimations. However, this could lead to a\nbias transfer towards the face quality assessment leading to discriminatory\neffects e.g. during enrolment. In this work, we present an in-depth analysis of\nthe correlation between bias in face recognition and face quality assessment.\nExperiments were conducted on two publicly available datasets captured under\ncontrolled and uncontrolled circumstances with two popular face embeddings. We\nevaluated four state-of-the-art solutions for face quality assessment towards\nbiases to pose, ethnicity, and age. The experiments showed that the face\nquality assessment solutions assign significantly lower quality values towards\nsubgroups affected by the recognition bias demonstrating that these approaches\nare biased as well. This raises ethical questions towards fairness and\ndiscrimination which future works have to address.\n