الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 157 |  |  |
| | | from | 157 | | |
Abstract
Biometric Presentation Attack Detection (PAD) is gaining increasing attention, especially with the wide use of smartphones in many applications regarding security and authentication. Users of mobile devices find it more convenient to unlock their smart applications with finger, face or iris recognition instead of passwords. On the other side, videos and photos of users are becoming more available online making it easier for attackers to spoof the authentication systems which rely on face and eye-region data for instance. Hence, it has become important to develop robust and generic techniques that protect against any presentation attack.
One major problem with current PAD systems is their lack of generalization to data captured by different sensors or in different environments. Conventional anti-spoofing methods assume that testing is from the same domain used for training, and so cannot generalize well on unseen attack scenarios. In light of this, we propose two end-to-end learning frameworks based on unsupervised Domain Adaptation (DA) to improve PAD generalization capability. Labeled source-domain samples are used to train the feature extractor and classifier via cross-entropy loss, while unsupervised data from the target domain are utilized in adversarial DA approach causing the model to learn domain-invariant features. Our first model is composed of symmetric classifiers and two per-class domain discriminators. Interaction between class probabilities and domain classification is utilized to jointly train a mobile-oriented feature extraction network, capable of generating domain-invariant features.
Using DA alone in face PAD fails to adapt well to target domain that is acquired in different conditions with different devices and attack types than the source domain. And so, in order to keep the intrinsic properties of the target domain, deep clustering of target samples is performed for our second proposed framework. Training and deep clustering are performed end-to-end, and experiments performed on several public benchmark datasets validate that our proposed Deep Clustering guided unsupervised Domain Adaptation (DCDA) can learn more generalized information compared with the state-of-the-art classification error on the target domain. The approaches are evaluated on six publicly available benchmark datasets (3 face datasets and 3 iris datasets) and cross-dataset testing is performed. We also performed cross-dataset testing on iris PAD datasets in terms of equal error rate which was not reported in literature before. Results show the effectiveness of the fine-tuned deep models in learning discriminative features that can tell apart bona-fide (real) from attack (fake) biometric images with very low error rate. Results of our proposed DA-based models show up to 40% improvement in cross-dataset Average Classification Error Rate (ACER) proving the effectiveness of these approaches in increasing the robustness and generalization of biometric PAD systems.