Can Your Face Detector Do Anti-spoofing?

Typical face recognition systems can be fooled by attacks as simple as a printed photograph or by playing a video recording in front of the camera. The attacks could be more sophisticated with 3D masks and involved makeup. Presentation attack detection (PAD) methods try to protect face recognition systems from these vulnerabilities, by having a separate pipeline to classify the liveness of face. The attacks can be of two types, impersonation attacks when somebody wants to be recognized as somebody else and obfuscation attacks where somebody wants to evade recognition. Most of the PAD systems involve a pipeline involving face detection, followed by some preprocessing and then the classifier which classifies the face image as real or spoof.