2020/07/22 by Anjith George, Sébastien Marcel, George, Anjith +1 · 1 citation
Computer Science · #Biometric Identification and Security #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences
paper · pdf · doi:10.48550/arxiv.2007.11457
openalex publication_date 2020/07/22 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Face recognition has evolved as a widely used biometric modality. However,\nits vulnerability against presentation attacks poses a significant security\nthreat. Though presentation attack detection (PAD) methods try to address this\nissue, they often fail in generalizing to unseen attacks. In this work, we\npropose a new framework for PAD using a one-class classifier, where the\nrepresentation used is learned with a Multi-Channel Convolutional Neural\nNetwork (MCCNN). A novel loss function is introduced, which forces the network\nto learn a compact embedding for bonafide class while being far from the\nrepresentation of attacks. A one-class Gaussian Mixture Model is used on top of\nthese embeddings for the PAD task. The proposed framework introduces a novel\napproach to learn a robust PAD system from bonafide and available (known)\nattack classes. This is particularly important as collecting bonafide data and\nsimpler attacks are much easier than collecting a wide variety of expensive\nattacks. The proposed system is evaluated on the publicly available WMCA\nmulti-channel face PAD database, which contains a wide variety of 2D and 3D\nattacks. Further, we have performed experiments with MLFP and SiW-M datasets\nusing RGB channels only. Superior performance in unseen attack protocols shows\nthe effectiveness of the proposed approach. Software, data, and protocols to\nreproduce the results are made available publicly.\n