A robust active-based face spoof detection for face recognition systems

With increasing rate of deployment of face recognition systems in various real life cases and applications, efforts by attackers is also on the rise with variety of spoofing approaches emerging. Hence, the development of a robust spoof detection technique is crucial. Although active-based techniques have proven robust in the task of spoof detection, they are faced with the problem of intrusiveness, high cost, computational complexity, low generalization ability and most of which require extra hardware. This study proposed an active-based robust spoof detection technique capable of detecting diverse forms of media or 2D attacks, yet, less intrusive, less expensive, low complexity, higher generalization ability than other active-based approaches. It does not require extra hardware and so can easily fit into existing systems. Spoof detection is achieved by analyzing the distortion changes of Video frames of user's face captured at different distances from the camera. Real-world facial photo and video data were also collected and used to generate both the legitimate and spoof attacks dataset. Utilizing both machine learning classifiers and deep learning model, the proposed approach achieved spoof detection with accuracy as high as 98.18%, with EER and HTER as low as 0.23 and 0.021 respectively.