Face Anti-Spoofing Based on NIR Photos

Face recognition technology is becoming more and more mature, which brings us great convenience in our life. However, the traditional RGB image has obvious shortcomings in Security check. The face of legitimate users is easily attacked by photos, videos and 3D masks. Therefore, it is very difficult to distinguish true and false faces only by RGB images. As a consequence, there is no denying that face anti-spoofing is very essential. In this paper, a binocular camera (RGB + NIR) is used for test. First, a RGB + NIR face database is creatively established. Because NIR cameras are naturally defensive against electronic devices and NIR information is more conducive to face anti-spoofing than RGB information, we extract NIR information and use feature vectors for binary classification. The results show that we proposed scheme had an excellent outcome. In this paper, the trained model is also applied to real live scenarios. And we also achieve a excellent progress.