Grad-CAM Applied to the Detection of Instruments Used in Facial Presentation Attacks
摘要
Biometric recognition, especially facial recognition, has achieved significant success, but it faces challenges like counterfeiting biometric data. This paper proposes a Facial Presentation Attack Detection (PAD) system that incorporates contextual information to identify and discard attacks involving detectable Presentation Attack Instruments (PAIs). The aim is to streamline computational efforts and enhance the subsequent PAD system’s analysis of facial features. The PAD system yields excellent results, achieving a 99% accuracy rate. This high performance is confirmed through the application of a Explainable Artificial Intelligence (XAI) technique, Grad-CAM.