A Deep Learning Technique to Countermeasure Video-Based Presentation Attacks

Presentation attack detection (PAD) on faces is crucial to countermeasure face recognition system from a security breach. The increase in convenience to our lives is not without its own additional avenue for exposure to a security breach. The presentation attack (PA), or spoofing attack is the act of using artificial/synthetic materials to get unauthorized access into a secured system. For example, an individual could use a 3D printed mask, or even a digital or printed photograph to circumvent a facial recognition authentication system, for the case of iris-detection an intruder could use custom contact lenses, and for fingerprints, digital prosthetics to penetrate a secured system. Previously, many deep learning approaches utilize a very deep complex model to handle face PAD. To countermeasure the above issue, in this paper, we apply a deep learning approach to mitigate the presentation attack. In our proposed approach, we implement a lightweight ‘modified-AlexNet’ and obtained the highest test accuracy of 99.89% on the Spoof in the Wild (SiW) dataset.