Conference Report on IEEE BIOSIG 2021
The 20th edition of the International Conference of the Biometrics Special Interest Group (BIOSIG) took place virtually at Fraunhofer IGD in Darmstadt, Germany from September 15 to 17. For BIOSIG-2021 an increased number of 114 individuals have participated. This scientific conference was the last part of the Darmstadt Biometric Week.
Christoph Busch welcomed the participants of the 20th conference edition with a brief history of BIOSIG and some of its highlights since its inception. The BIOSIG conference and the community around it facilitated or directly contributed to several key developments in biometrics, including ISO/IEC standards and the foundation of the European Association for Biometrics. Following the short welcome from the General Chairs, Javier Galbally (European Commission, Joint Research Centre) delivered an entertaining presentation highlighting scientific papers with remarkable research topics and catchy titles.
The technical part of the BIOSIG-2021 conference featured 16 oral presentations and 16 poster presentations, as well as 3 keynotes by Istvan Szilard Racz (EU-Lisa), Richa Singh (IIT Jodhpur), and Brendan Klare (Rank One Computing).
In the first keynote, Istvan Szilard Racz (EU-Lisa) presented the current state of European border management along with changes and extensions which are under development or moving into operation. The talk concentrated on the Entry/Exit System (EES), which will register biographic and biometric (fingerprints and face) information of travellers from third-countries each time an external EU border is crossed. EES will facilitate interconnectedness of existing infrastructure (i.e. EURODAC, SIS, VIS) and efficient exchange of information. Interoperability with legacy data and across different acquisition methods, sample qualities, and sensor types is challenging and critically important for the efficacy of the overall system. This topic has also been addressed in one of the presentations during the conference by Brady Williams (West Virginia University), who discussed interoperability for contact and contactless fingerprint sensors. The EES infrastructure is designed with scalability in mind: it is able to handle hundreds of million of records, implements strict data and biometric sample, while biometric performance targets are mandated by the legal base and tested continually. In the future, connections to other European systems and databases (biometric and otherwise) are envisioned within the scope of the framework for interoperability between EU information systems in the field of borders and visa and in the field of police and judicial cooperation, asylum and migration. The key goals of the EES are protection of EU citizens’ security, prevention of irregular migration, streamlining the border control processes, tracking of visa overstays, and detection of identity fraud. EES will operate in full respect of fundamental rights and data protection. Such strong legal base and privacy-preservation are indispensable for operational biometric systems. The presentation of Jan Czarnocki (KU Leuven Centre for IT & IP Law) provided further insights into legal and ethical issues and challenges in the context of the proposed EU Artificial Intelligence Act, while Tamas Bisztray (University of Oslo) highlighted potential privacy risks which may arise from loopholes in the General Data Protection Regulation. Both presentations emphasised the need for precise legal terminology and definitions.
Several talks addressed a fundamental and perhaps insufficiently popular topic of biometric performance evaluation metrics and protocols. James Wayman (Office of Biometric Identity Management, Dept. Of Homeland Security) talked about evaluation protocols and consequences of biometric scenarios where the enrolment database contains multiple samples for each identity. Teodors Eglitis (Roma TRE University) also addressed the topic of experimental protocols and highlighted their importance in biometric evaluations. Well-defined and practically reasonable experimental protocols are vital for benchmarks of recognition algorithms and reproducible research. Shigefumi Yamada (Japan Automatic Identification Systems Association) presented a statistical method for modelling and estimating false match rates in biometric recognition systems. The method is particularly interesting for scenarios where a sufficiently large sample of real data is otherwise not available. The insights and contributions of those three presentations are expected to be included in the ongoing biometric standardisation efforts of ISO/IEC.
Richa Singh (IIT Jodhpur) opened the second day of the conference with a keynote on dependable biometric systems. The keynote concentrated on two topics – adversarial attacks and fairness. Currently, many systems, algorithms, and databases are somehow biased (e.g. w.r.t. demographic attributes) and not robust against attacks. Two specific high-visibility examples like biased online search and high-resolution deepfake videos served to emphasise this point. A further challenge in this context is explainability of algorithms, i.e. understanding why certain outputs are generated. Bias and attacks can undermine the operational efficacy and public trust in biometric and other automated systems. The researchers and organisations are therefore actively developing methods which can detect and/or mitigate those issues. However, simultaneously achieving generalisability across different data sources, detection and attack models, and attack types is highly challenging and as of yet remains an unsolved problem. The overall research goal can be concisely summarised as development of bias-invariant, generalised, unified algorithms for detection of deepfakes and other attacks. While promising advances have been made in recent years, many challenges and open questions still remain; furthermore, the attacks likewise keep on improving. A couple of other speakers addressed related topics in their presentations. Kazuya Kakizaki (NEC Corporation & University of Tsukuba), Zohra Rezgui (University of Twente), and Inderjeet Singh (NEC Corporation) talked about adversarial attacks on face recognition and classification; conversely, Tomoaki Matsunami (Fujitsu Limited) and Alperen Kantarci (Istanbul Technical University) presented methods of presentation attack detection.
The talks of John McCauley (West Virginia University) and Christian Rathgeb (Hochschule Darmstadt) pertained to individuals who may inadvertently pose significant challenges to biometric systems merely by their visual appearance and not by malicious intent. In particular, they talked about identical twins and so-called “doppelgängers” who exhibit highly similar facial appearances. Although such individuals account for a small fraction of the population, reliably recognising them by automated algorithms is difficult and possibly represents an inherent limitation of facial recognition. The experiments in both publications relied on newly collected datasets which can be made available to other researchers upon request.
Biometric sample quality has been a prominent topic throughout the conference. The presentations and posters of Biying Fu (Fraunhofer IGD), João Tremoço (University of Coimbra), and Claudio Yáñez (TOC Biometrics) addressed the topic of facial image quality assessment and applications of sample quality within the recognition or attack detection pipelines. Pedro Neto (INESC TEC & Universidade do Porto) and Mustafa Ekrem (Istanbul Technical University) investigated facial recognition on occluded and masked faces; the latter introduced a dataset which is now publicly available. Face recognition in the presence of facial masks is an especially relevant topic in light of the ongoing COVID-19 pandemic, which affected biometric technologies in numerous ways.
Sample quality and unconstrained recognition in biometric modalities beyond the face was also addressed by multiple presenters. Tim Oblak (University of Ljubljana) presented an open-source toolbox for fingermark quality assessment, while Michael Linortner (University of Salzburg) investigated the relevant of certain properties of vascular patterns for vein recognition performance. Further research on vascular biometrics has been prominently featured in the poster session. It included presentations by Victor Bros (Idiap), Bernhard Prommegger (University of Salzburg), and Felix Marattukalam (University of Auckland) concentrated on the applications of various deep learning methods to vascular recognition, as well as an open-source framework for vein recognition presented by Michael Linortner (University of Salzburg).
The aforementioned impact of COVID-19 pandemic on facial recognition was discussed in-depth in the last keynote of the conference by Brendan Klare (Rank One Computing). The talk highlighted key developments within facial recognition pre-pandemic: following massive gains in biometric performance due to advances in deep learning, availability of large training datasets, and extensive computational resources, the prevalence and investment into facial recognition has increased substantially. Both security and convenience-oriented systems benefit from face recognition, with operational applications being used by governmental agencies, commercial entities, and individual customers. Demographic bias has been a hot topic in the context of facial recognition; the talk simultaneously emphasised the need for fair recognition systems as well as the examples of misperceptions in media and society in this area. From the academic and benchmarking perspective, clean and reliable datasets and their corresponding groundtruth labels are critically important to achieve meaningful experimental results and conclusions. In other words, data integrity is a key issue in this context. The restriction caused by the COVID-19 pandemic have led to an explosive growth of (remote) digital identity verification services. Furthermore, hygienic requirements have necessitated facial recognition algorithms which work reliably in the presence of facial masks. The industry has adapted to those requirements remarkably quickly; other modalities, such as iris and contactless fingerprint recognition might also gain relevance in this context. The keynote also heavily featured topics related to security and privacy of biometric data, advocating for a privacy-by-design approach. Those topics were also addressed in other talks during the conference in presentations by Hiroto Tamiya (NEC Corporation), who talked about post-quantum security of biometric template protection systems, and by Amina Bassit (University of Twente), who investigated and compared Bloom Filter-based and homomorphic encryption-based template protection schemes with respect to the fulfilment of ISO/IEC 24745 (and other) requirements for biometric template protection. The talk of Xingbo Dong (Monash University) demonstrated the relevance of biometric template protection by showing the feasibility of generation of high-definition facial images from templates containing features extracted by deep neural networks.
In addition to the aforementioned keynotes and topic clusters, the conference featured a wide variety of presentations on other topics. Applications of neural networks for recognition of humans based on their gait patterns and robust speaker recognition were presented by Enrique Argones Rua (KU Leuven) and Labib Chowdhury (North South University), respectively. Parisa Rezaee Borj (NTNU) presented a method for detection of sexual predators in chats.
The social event included a funny quiz in which everybody could test their knowledge about the history of BIOSIG publications. It was moderated by Christoph Busch and won by Olaf Henniger to receive the “Best Observer Award”. As in previous editions of the BIOSIG conference, participants of the conference themselves voted for the best paper and the best poster presented at the conference. The BIOSIG 2021 “Best Paper Award” was given to Amina Bassit (University of Twente) for the interesting presentation “Bloom Filter vs. Homomorphic Encryption: Which approach protects the biometric data and satisfies ISO/IEC 24745?“. The winner of the “Best Poster Award” was Biying Fu (Fraunhofer IGD) for the poster “The effect of face morphing on face image quality”.
The BIOSIG conference was preceded by several satellite events in the Darmstadt Biometric Week:
- EAB Research Projects Conference (EAB-RPC) 2021
- EAB General Assembly
- TeleTrusT Biometric Working Group Meeting
- EAB Biometrics Max Snijder, Research, and Industry Awards 2021
- EAB Biometrics Training Event
The 2021 BIOSIG conference was jointly organized by the Competence Center for Applied Security Technology (CAST) and the special interest group BIOSIG of the Gesellschaft für Informatik e.V. (GI).
The conference was technically co-sponsored by IEEE Biometric Council and the papers will be added to IEEE Xplore.
Next year, the BIOSIG conference is planned to take place between September 14 to 16, 2022 in Darmstadt, Germany.