Guest Editorial: BIOSIG 2020 special issue on trustworthiness of person authentication

Recent guidelines for ‘Trustworthy AI’ state that it not only relates the trustworthiness of the AI system itself but also comprises the trustworthiness of all processes and actors that are part of the system's life cycle. Person authentication is a particular application of AI in which (i) the compliance to laws and regulations; (ii) the respect for ethical principal and values; (iii) and the robustness, both from a technical and social perspective, are of crucial importance.

This is the first IET Biometrics ‘Trustworthiness of Person Authentication’ special issue, having as starting point the 2020 edition of the Biometric Special Interest Group (BIOSIG) conference. This special issue gathers works focussing on topics of biometric recognition put under the new light of fostering the trustworthiness of the involved processes.

The ‘BIOSIG 2020 special issue on Trustworthiness of Person’ issue contains seven papers, most of them being extended versions of papers presented at the BIOSIG 2020 conference, dealing with concrete research areas within biometrics such as presentation attack detection (PAD), traditional and emergent biometric characteristics, and biometric recognition and soft biometrics in the presence of facial masks.

The paper ‘Unknown Presentation Attack Detection against Rational Attackers’, by Ali Khodabakhsh and Zahid Akhtar, investigates the vulnerability of PAD systems to attacks in real-life settings, addressing the detection of unknown attacks, the performance in adversarial settings, few-shot learning, and explainability. In this study, these limitations are addressed through an approach that relies on a game theoretic view for modelling the interactions between the attacker and the detector. These challenges are successfully addressed, and the methodology proposed provides a more balanced performance across known and unknown attacks, achieving at the same time state-of-the-art performance in known and unknown attack detection cases against rational attackers. Lastly, the few-shot learning potential of the proposed approach is studied as well as its ability to provide pixel-level explainability.

The paper ‘On the Generalisation Capabilities of Fisher Vector based Face Presentation Attack Detection’ by Lazaro Gonzalez-Soler, Marta Gomez-Barrero and Christoph Busch, focusses on face PAD in more challenging scenarios, where unknown attacks are included in the test set. Considering those more realistic scenarios, in which the existing algorithms face difficulties in detecting unknown presentation attack instruments (PAI), the authors propose a new feature space based on Fisher vectors, computed from compact binarised statistical image features' (BSIF) histograms, which allow discovering semantic feature subsets from known samples in order to enhance the detection of unknown attacks. This new representation, evaluated for challenging unknown attacks taken from freely available facial databases, shows promising results in the presence of unknown attacks. Furthermore, the proposed methodology achieves state-of-the-art performance in cross-dataset scenarios.

The paper ‘Failure of Affine-based Reconstruction Attack in Regenerating Vascular Feature Points’ by Mahshid Sadeghpour, Arathi Arakala, Stephen Davis and Kathy Horadam, focusses on the vulnerabilities of biometric systems based on retina and hand vascular data to inverse biometrics attacks. In particular, affine-based reconstruction attack methods, modelling the biometric recognition algorithm by an affine approximation, are considered. This type of attack reconstructs targetted biometric references using the modelled biometric recognition algorithm and the comparison scores issued by the system. Even though this reconstruction method has only been successfully applied to reconstruct face images, the common consensus is that any biometric system that issues comparison scores could be vulnerable to such an attack, since the method is sufficiently general to be applied to other biometric templates. In this work, the authors show that the attack fails to regenerate sparse vascular feature point templates in experiments that test the reconstruction attack on feature point patterns extracted from retina and hand vascular images. The experimental results show that the reconstruction attack is not as threatening as it is widely accepted to be and that vascular biometric template protection schemes that store sparse templates as references and reveal comparison scores are not susceptible to affine-based reconstruction attacks.

The paper ‘CNN based Off-angle Iris Segmentation and Recognition’, by Ehsaneddin Jalilian, Mahmut Karakaya and Andreas Uhl, examines thoroughly the general effect of different gaze angles on ocular biometrics and then relates the findings to the CNN-based off-angle iris segmentation results and the subsequent recognition performance. While deep learning techniques (i.e., segmentation-based CNNs) are increasingly used to address this problem, a significant lack of information about the mechanism affecting the related distortions on the performance of these networks remains. Specifically, there is a need for a comprehensive recognition framework dedicated to specific off-angle iris recognition using such modules. The authors introduce an improvement scheme to compensate for some segmentation degradations caused by the off-angle distortions and further propose a new gaze-angle estimation and parameterisation module to estimate and rectify the off-angle iris images back to frontal view. Taking benefit of these, the authors formulate several approaches to configure an end-to-end framework for the CNN-based off-angle iris segmentation and recognition.

The paper ‘Subject Independent Evaluation of Eyebrows as a Stand-alone Biometric’, by Hoang Nguyen, Ajita Rattani and Reza Derakhshani, explores emergent ocular modalities to address challenges such as occlusions due to face coverings. The authors present ocular biometrics that use features extracted from the eye region and around it, such as the eyebrow region, as a potential remedy for these challenges. This work evaluates five deep learning models (lightCNN, ResNet, DenseNet, MobileNetV2, and SqueezeNet) for eyebrow-based user authentication in a subject-independent environment across different datasets, lighting conditions, resolutions, and facial expressions. The authors also present a challenging simulated identical twins scenario in the training and testing datasets as well as results obtained using two well-known databases (FACES and VISOB).

The papers ‘An Extended Evaluation of the Effect of Real and Simulated Masks on Face Recognition Performance’, by Naser Damer, Fadi Boutros, Marius Süßmilch, Florian Kirchbuchner and Arjan Kuijper and ‘Facial Masks and Soft-Biometrics: Leveraging Face Recognition CNNs for Age and Gender Prediction on Mobile Ocular Images’, by Fernando Alonso-Fernandez, Kevin Hernandez-Diaz, Silvia Ramis, Francisco J. Perales and Josef Bigun, both address challenges posed by COVID-19 face masks on biometric recognition.

In the first paper, the authors present a specifically collected database containing three sessions under different capture conditions to simulate realistic use cases and additionally perform data augmentation to include synthetic mask occlusions. The paper studies the effect of masked face probes on the behaviour of four face recognition systems (academic and commercial) and performs an evaluation including masked to non-masked and masked to masked face comparisons. Furthermore, the work presents a comparison of the effect of real masks versus the simulated masks on face recognition performance.

The second paper, address the use of selfie ocular images captured with smartphones to estimate age and gender in the scenario of partial face occlusions due to the mandatory use of face masks. This work explores the challenges posed by the explosion of the use of mobile devices and increased migration to digital services caused by the pandemic. In particular, due to mobile devices' hardware limitations and size restrictions of downloadable applications, it is infeasible to employ large CNNs in tasks such as identity or expression recognition. Thus, the authors adapt two existing lightweight CNNs proposed in the context of the ImageNet Challenge and two additional architectures proposed for mobile face recognition. The overfitting problem is addressed by using networks pre-trained on ImageNet and some networks that are further fine-tuned for face recognition, for which very large training databases are available. Since both tasks employ similar input data, the authors hypothesise that the proposed strategy can be beneficial for soft-biometrics estimation. A comprehensive study of the effects of different pre-training over the employed architectures is carried out, showing that, in most cases, a better accuracy is obtained after the networks have been fine-tuned for face recognition.