JIITA, vol.6 no.3, p.610-616, 2022, DOI: 10.22664/ISITA.2021.6.3.610
Myoung-Kyu Sohn 1,*), Hyunduk Kim 1)
1) Division of Automotive Technology, DGIST, Daegu, Republic of Korea
Abstract: As face recognition technology has become more common, image-based face authentification systems are being widely used. At the same time, techniques for hacking these technologies have begun to appear. Face spoofing, which attempts to attack the face recognition system by showing a printed face image, an on-screen image, or a 3D printed mask, etc. instead of the real thing, is a major problem in the face recognition system. In addition, as the technology of image recognition is advanced, the anti-spoofing method using it is gradually improving. 3D cameras or various expensive sensors may be used, but this has a problem of poor actual commercialization and scalability. To solve this problem, this paper intends to develop an anti-spoofing system using a single RGB sensor using deep learning-based recognition technology. It also proposes a system that operates in real-time by using a lighter model than the existing complex algorithm
Keywords: deep learning, real-time face recognition, face anti-spoofing, presentation attack, face liveness detection
Fullpaper: