Face presentation attack detection based on a statistical model of image noise

Abstract : The vulnerability of most existing face recognition and authentication systems against face presentation attacks (a.k.a. face spoofing attacks) has been mentioned and studied in many works. This paper introduces a novel parametric approach for face PAD using a statistical model of image noise. In fact, facial images from a presentation attack contain specific textural information caused by the presentation process which makes them different from bona-fide images. The subtle difference between bona-fide and presentation attack images can be interpreted by the difference regarding noise statistics within the skin zone of the face. Our solution is casted in the hypothesis testing framework. A new database for face PAD containing face bona-fide images and images of high-quality presentation attacks has been also introduced. The performance of the proposed approach was proven in the mentioned database. Experimental results show that, in a controlled situation, our solution performs better than the other approaches in the literature.