Real-Time 3D Texture and Motion Analysis for Face Anti-spoofing Using Deep Learning and Computer Vision

Along with the rise of face recognition systems, there is a growing concern over the vulnerability of these systems to spoofing attacks, where an attacker can present a fake or a manipulated face image to bypass the authentication process. Regular updates in the anti-spoofing systems are required to stop attacks on the recognition systems. The existing systems use convolution neural networks as binary classifiers to differentiate between a real face and a spoofed face. We propose a face anti-spoofing system that aims to differentiate between a live and spoofed face image by employing a two levels of verification and. The proposed system is designed to work with RGB cameras and does not require any 3D camera hardware or stereo vision cameras, making it accessible for more practical applications. The proposed system uses a transfer learning technique to perform first level of verification and validate in the second level using computer vision. Computer vision aims to build a high-security model for cutting out all the attacks on the facial recognition system.