Fake Face Image Detection Using Deep Learning-Based Local and Global Matching

Abstract

The widespread adoption of face recognition systems in practice has provoked multiple attempts to fail these systems in order to impersonate another person. The range of such fake attacks is wide, and methods which can be used to compensate for one type of attacks are not adapted against other attacks. In this study, we propose a method for detecting fake face images based on local and global matching provided by deep neural networks. Also we do not discard the background analysis as a pre-processing stage. The idea is to assess the depth of the face in a still image as one of the main features of liveliness, which is not an easy task. The proposed method is directed against presentation attacks and attacks of adversarial perturbations. The experiments were conducted with and without deep neural networks. The use of deep learning increased the true accept rate and significantly reduced the error values.