A Self-explainable Face Anti-spoofing Solution Based on Depth Estimation

The human face is one of the most widely available biometric methods of identification and verification. In the age of Industry 4.0, one can find digital cameras everywhere, making a face recognition-based digital identity system much more viable. The face is vulnerable to spoofing attacks because it is the most accessible and commonly used biometric information among all biometric modalities. Most deep learning-based face anti-spoofing solutions have poor generality, and none of them provides a rationale for their results. We propose an explainable AI model that not only classifies spoof and live faces but also explains why they are different using mid-level features. Our proposed end-to-end face anti-spoofing network extracts the depth map and classifies the face input.