Unravelling Robustness of Deep Face Recognition Networks Against Illicit Drug Abuse Images

Abstract

Alteration in facial features can lead to a significant drop in recognition performance. These alterations can be due to several factors: one such prominent and less explored factor is illicit drug abuse. To advance the understanding of how drug abuse faces affect the performance of state-of-the-art deep face recognition (DFR) networks, in this study, we have utilized clean and illicit drug abuse faces. Extensive studies are performed on deep face recognition and soft biometric identification, such as gender, ethnicity, and expression recognition. It is observed that illicit drug abuse not only impacts the identity recognition performance but also degrades the soft biometrics identification accuracy. Therefore, to advance the integrity of DFR, we have performed the detection of illicit drug abuse as a potential solution to its mitigation. In the end, the robustness of the drug abuse face detector is evaluated under the prominent use of social-media filters on face images.