Two-Stage Face Detection and Anti-spoofing

Face recognition is a widely used biometric technique that has received a lot of attention. It is used to establish and verify the user's identity, and subsequently grant access for authorized users to restricted places and electronic devices. However, one of the challenges is face spoofing or presentation attack allowing fraudsters who attempt to impersonate a targeted victim by fabricating his/her facial biometric data, e.g., by presenting a photograph, a video, or a mask of the targeted person. Several approaches have been proposed to counteract face spoofing known as face anti-spoofing techniques. This paper's major goals are to examine pertinent literature, and develop and evaluate a two-stage approach for face detection and anti-spoofing. In the first stage, a multi-task cascaded convolutional neural network is used to detect the face region, and in the second stage, a multi-head attention-based transformer is used to detect spoofed faces. On two benchmarking datasets, a number of experiments are carried out and examined to assess the proposed solution. The results are encouraging, with a very high accuracy, which encourages further research in this direction to build more robust face authentication systems.