Hidden adversarial attack on facial biometrics - a comprehensive survey
Sheilla Ann Bangoy Pacheco, Jheanel Espiritu Estrada, Mahesh M Goyani · Procedia Computer Science · 2025
Deep learning models are widely used in pattern recognition and computer vision applications. Such models are susceptible to adversarial attacks like blur, noise, translation, flip, rotation, illumination change, etc. Their vulnerability to adversarial attacks poses critical challenges in terms of privacy, security, and reliability. Attackers target such systems and manipulate input or the model parameters to misbehave the model. Such attacks are big threats to authorization systems, such as criminal identification, attendance monitoring, surveillance, airport, and human-computer interaction (HCI) systems. Face recognition is one of the most widely used biometric systems and could be a soft target for attackers. This article presents a comprehensive study of adversarial attacks on face recognition systems, their implications, threat models, etc. By understanding the nature of adversarial threats and adopting robust defense strategies, can pave the way for safer and more trustworthy face recognition systems. This article summarized state-of-the-art adversarial attacks with their characteristics. The article also through lights on open research areas and the future scopes in the adversarial attack. This comprehensive review can serve as a ready-to-go reference for readers.