Robust 2D/3D face mask presentation attack detection scheme by exploring multiple features and comparison score level fusion

The face mask presentation attack introduces a greater threat to the face recognition system. With the evolving technology in generating both 2D and 3D masks in a more sophisticated, realistic and cost effective manner encloses the face recognition system to more challenging vulnerabilities. In this paper, we present a novel Presentation Attack Detection (PAD) scheme that explores both global (i.e. face) and local (i.e. periocular or eye) region to accurately identify the presence of both 2D and 3D face masks. The proposed PAD algorithm is based on both Binarized Statistical Image Features (BSIF) and Local Binary Patterns (LBP) that can capture a prominent micro-texture features. The linear Support Vector Machine (SVM) is then trained independently on these two features that are applied on both local and global region to obtain the comparison scores. We then combine these scores using the weighted sum rule before making the decision about a normal (or real or live) or an artefact (or spoof) face. Extensive experiments are carried out on two publicly available databases for 2D and 3D face masks namely: CASIA face spoof database and 3DMAD shows the efficacy of the proposed scheme when compared with well-established state-of-the-art techniques.