Can GAN-induced Attribute Manipulations Impact Face Recognition?

TL;DR

This paper investigates how GAN-generated modifications of facial attributes, such as eyeglasses and sex cues, affect the accuracy of face recognition systems, revealing significant performance impairments.

Contribution

It is the first study to analyze the impact of GAN-induced attribute manipulations on face recognition performance.

Findings

Attribute manipulations can impair face recognition accuracy by up to 73%.

Eyeglasses and sex cue alterations have the most significant impact.

GAN-based attribute modifications pose challenges for face recognition robustness.

Abstract

Impact due to demographic factors such as age, sex, race, etc., has been studied extensively in automated face recognition systems. However, the impact of \textit{digitally modified} demographic and facial attributes on face recognition is relatively under-explored. In this work, we study the effect of attribute manipulations induced via generative adversarial networks (GANs) on face recognition performance. We conduct experiments on the CelebA dataset by intentionally modifying thirteen attributes using AttGAN and STGAN and evaluating their impact on two deep learning-based face verification methods, ArcFace and VGGFace. Our findings indicate that some attribute manipulations involving eyeglasses and digital alteration of sex cues can significantly impair face recognition by up to 73% and need further analysis.