On Effectiveness of Anomaly Detection Approaches against Unseen Presentation Attacks in Face Anti-spoofing

Abstract

While face recognition systems got a significant boost in terms of recognition performance in recent years, they are known to be vulnerable to presentation attacks. Up to date, most of the research in the field of face anti-spoofing or presentation attack detection was considered as a twoclass classification task: features of bona-fide samples versus features coming from spoofing attempts. The main focus has been on boosting the anti-spoofing performance for databases with identical types of attacks across both training and evaluation subsets. However, in realistic applications the types of attacks are likely to be unknown, potentially occupying a broad space in the feature domain. Therefore, a failure to generalize on unseen types of attacks is one of the main potential challenges in existing anti-spoofing approaches. First, to demonstrate the generalization issues of two-class anti-spoofing systems we establish new evaluation protocols for existing publicly available databases. Second, to unite the data collection efforts of various institutions we introduce a challenging Aggregated database composed of 3 publicly available datasets: Replay-Attack, Replay-Mobile and MSU MFSD, reporting the performance on it. Third, considering existing limitations we propose a number of systems approaching a task of presentation attack detection as an anomaly detection, or a one-class classification problem, using only bona-fide features in the training stage. Using less training data, hence requiring less effort in the data collection, the introduced approach demonstrates a better generalization properties against previously unseen types of attacks on the proposed Aggregated database.

Biometric Face Presentation Attack Detection With Multi-Channel Convolutional Neural Network

“…Since a new database is proposed in this paper, baseline experiments are performed for each channel first. The selected baselines are reproducible and have either open-source implementation [12] or re-implementation. Three sets of baselines are described in this section.…”

“…The source codes for the network and instructions to train the model are made publicly available allowing to reproduce the findings. We benchmark the proposed method against selected reproducible baseline available in recent publications on the topic [12], as well as reimplementations of recent literature in multi-channel PAD [13]. We demonstrate that the multi-channel approach is beneficial in both proposed and baseline systems.…”

Pedagogical approaches for e‐assessment with authentication and authorship verification in Higher Education

“…Pilot coordinators play an important role to communicate needs and technical issues to the technology developers and keep course teams and students-including special educational needs students-well informed and supported on security and privacy as well as on fraud detection (QAA, Vol 50 No 6 2019, prevention and trust (Mellar, Peytcheva-Forsyth, Kocdar, Karadeniz, & Yovkova, 2018;Park, 2004).…”

“…Good academic practice in (online) teaching should recognise academic malpractice such as external and intrinsic plagiarism (Appelgren Heyman et al ., ; Ferrell, ; Potthast, Eiselt, Barrón Cedeño, Stein, & Rosso, ) including impersonation and presentation attack (Nikisins, Mohammadi, Anjos, & Marcel, ), which has become more common. All these forms of cheating and plagiarism should be addressed by academic staff (Ivanova et al ., ; Kambourakis & Damopoulos ; Pell, ) with pedagogical strategies for e‐assessment supported by the effective use of technologies, such as the TesLA system.…”

Deep Pixel-wise Binary Supervision for Face Presentation Attack Detection

“…Image quality measures [9], and image distortion analysis [24] use the deterioration of the sample quality and artifacts in the re-capture as a cue for presentation attack detection. Most of these methods treat PAD as a binary classification problem which reduces its generalization capability in an unseen attack scenario [19]. Nikisins et al [19] proposed a framework for one class classification using one class Gaussian Mixture Models (GMM).…”