Biometric Presentation Attack Detection using Gaze Alignment

Face recognition systems have been improved rapidly in recent decades. However, their wide deployment has been hindered by their vulnerability to spoofing attacks. In this paper, we present a challenge and response method to detect attack in face recognition systems by recording the gaze of a user in response to a moving stimulus. The proposed system extracts eye centres in the captured frames and computes features from these landmarks to ascertain whether the gaze aligns with the challenge trajectory in order to detect spoofing attacks. The system is tested using a new database simulating mobile device use with 70 subjects attempting three types of spoof attacks (projected photo, looking through a 2D mask or wearing a 3D mask). Evaluations on the collected database show that the proposed approach performs favourably when compared with state-of-the-art methods.