NEWSLETTER
Vol22-Newsletter
Vol22-Newsletter
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>NEWSLETTER</strong><br />
Volume 22, 2017<br />
Content<br />
Two New Vice President s<br />
IEEE Biometrics Council welcomes two new Vice Presidents in<br />
the Executive Committee. Dr. Sébastien Marcel and Prof. Pong C.<br />
Yuen will be Vice Presidents of Conferences and Technical<br />
Activities, respectively. The Council thank outgoing Vice<br />
Presidents Prof. Sudeep Sarkar and Prof. Ioannis Kakadiaris for<br />
their services and contributions.<br />
- New Vice Presidents<br />
- IEEE Trans. on Biometrics<br />
- Biometrics News<br />
- Biometrics Interview:<br />
Kenneth D. Gantt<br />
- Spotlight<br />
- Research Paper Summary<br />
- Conference Report:<br />
BIOSIG2016<br />
Prof. Yuen is a Professor and the Head of the Department of<br />
Computer Science. His research interests include video<br />
surveillance, human face recognition, and biometric security and<br />
privacy. Dr. Yuen serves as a Hong Kong Research Grant Council<br />
Engineering Panel Member. He is an Editorial Board Member of<br />
Pattern Recognition and an Associate Editor of IEEE T-IFS and<br />
SPIE Journal of Electronic Imaging. He was the Program Co-Chair<br />
of the IEEE Fifth International Conference on Biometrics: Theory,<br />
Applications and Systems in 2012.<br />
LOREM IPSUM DOLOR SIT AMET<br />
Dr. Sébastien Marcel is a Senior Research Scientist with the Idiap<br />
Research Institute, where he leads the Biometrics Group and<br />
conducts research on multimodal biometrics, including face<br />
recognition, speaker recognition, vein recognition, and spoofing<br />
and anti-spoofing. He is also the Director of the Swiss Center for<br />
Biometrics Research and Testing. He serves as an Associate<br />
Editor of the IEEE T-IFS. He is Coeditor of the upcoming<br />
Handbook on Biometric Anti-Spoofing (Springer), and an Area<br />
Editor of the Encyclopedia of Biometrics? Second Edition.<br />
- Calender
I EEE Transactions on Biometrics, Behavior, and I dentity Science<br />
Proposal on IEEE Transact ions on Biomet rics, Behavior, and Ident it y Science<br />
Biometrics and Identity Science is a core opportunity area with a large social impact which is becoming a<br />
major component in several large-scale identity management projects. In US, Europe, Asia, Africa, and<br />
Australasia, there are several active program funding biometrics-related activities. For example, OBIM,<br />
India?a UIDAI (Aadhaar) project, FBI?s NGI, and other nation-wide identity management programs have<br />
shown the usefulness of biometrics for identity management.<br />
IEEE Biometrics Council has submitted a proposal to IEEE for starting a new Transactions, tentatively named<br />
as "IEEE Transactions on Biometrics, Behavior, and Identity Science". The proposed IEEE Transactions on<br />
Biometrics, Behavior, and Identity Science would be a major source of dissemination of scientific<br />
innovations and progress in this domain. A broad scope would also ensure a widespread reach to academic,<br />
technical, engineering, and legal/law enforcement communities investing in biometrics. The proposed topics<br />
include,<br />
- Management of human identities using traditional behavioral and physiological characteristics<br />
(biometric modalities) including Face, Fingerprint, Iris, Periocular, Palmprints, Gait, Ear, and Voice.<br />
- Emerging Biometrics including Social and Behavioral Biometrics, Gesture based Biometrics,<br />
Keystroke, Vascular, Knuckle, and Soft Information/Biometrics such as Quality, Age, Gender, Kinship,<br />
Social Context, and Biographic Analysis<br />
- Pattern Classification and Machine Learning, including Deep Learning, Algorithms for Biometrics,<br />
Behavior and Identity Science<br />
- Multibiometrics including Fusion Approaches<br />
- Person Re-identification<br />
- Novel Biometric Sensors<br />
- Biometrics Anti-spoofing, Template Protection, Template Update, and Data Protection<br />
- Adversarial Attacks on Biometrics Systems<br />
- Detection, Tracking, Identification, and De-identification in Surveillance<br />
- Novel (challenging) Biometrics Databases, Protocol and Benchmarking, Large Scale Evaluation,<br />
Confidence Interval Estimation, and Performance Modeling and Prediction<br />
- Mobile Biometrics, Active Authentication, Mobile Device Security, and Device Identification<br />
- Forensic Analysis in Identity and Activity<br />
- Biometrics and Human Behavior Analysis, Human Activity Understanding, and Behavior Modeling<br />
- Modeling the Interplay of Behavioral Modalities with Applications in Biometrics and Security<br />
- Applications of Biometrics and Identity in Privacy Preserving Computing, User Centric Security,<br />
Usable Privacy and Security, Social and Criminal Network Inference, Privacy Protection and<br />
Enhancement<br />
- Applications of Biometrics and Identity in Modeling Deception, Deceptive Intent, and Predictive<br />
Analytics<br />
- Applications of Biometrics in Law Enforcement and Cybercrime, Finance, Smart-cards, Border<br />
Control, Healthcare, Civil Registry, Access Control, Securing Documents, Fraud Detection, Cloud, and<br />
Entertainment.<br />
The IEEE Biometrics Council request the research community to fill the online form to show their interest in<br />
the proposal Transactions: ht t ps:/ / t inyurl.com / km yqhl4<br />
Deadline to submit the online form: March 31, 2017<br />
If you have any feedback on this proposal, please send an email to Dr. Mayank Vatsa, Vice President<br />
(Publications) at mayank@iiitd.ac.in
Biometrics News<br />
Propriet ary Fingerprint Templat e Evaluat ion (PFTII)<br />
The NIST Proprietary Fingerprint Template Test Phase II is part of an ongoing program to measure the<br />
performance of fingerprint matching software by utilizing vendor proprietary fingerprint templates. The<br />
99.98% true match rate, combined with the related equal error rates achieved by Innovatrics?s algorithm,<br />
are the best results ever recorded during PFT evaluations.<br />
?This unique combination of industry-leading accuracy and top-ranked speed underlines our belief that<br />
Innovatrics?s algorithm is the most suitable solution for practical implementation and use with large-scale<br />
databases in Automatic Fingerprint Identification System (AFIS) solutions,? said Jan Lunter, CEO of<br />
Innovatrics. The test assesses the accuracy of end-stage matchers ? the computationally expensive<br />
algorithms used in the last stages of one-to-many AFIS searches.<br />
More: https://www.nist.gov/itl/iad/image-group/proprietary-fingerprint-template-evaluation-pftii<br />
St udy looks deeper int o face recognit ion<br />
Researchers at Carnegie Mellon University have released a new study that seeks to explain how our brains<br />
process facial images so quickly and accurately. The new study, published in the Dec. 26, 2016 issue of the<br />
Proceedings of the National Academy of Sciences (PNAS), examined how humans use their brains by using<br />
magnetoencephalography (MEG) scans.<br />
The team used highly sophisticated brain imaging tools and computational methods to measure the<br />
real-time brain processes that convert the appearance of a face into the recognition of an individual. "Our<br />
results provide a step toward understanding the stages of information processing that begin when an image<br />
of a face first enters a person's eye and unfold over the next few hundred milliseconds, until the person is<br />
able to recognize the identity of the face," said Mark D. Vida, a postdoctoral research fellow in the Dietrich<br />
College of Humanities and Social Sciences' Department of Psychology and Center for the Neural Basis of<br />
Cognition (CNBC).<br />
More: http://www.planetbiometrics.com/article-details/i/5367/<br />
Liveness Det ect ion Compet it ion Arrives at Crit ical Time for Iris Scanning<br />
With the technology thus poised to reach a huge number of consumers, the issue of liveness detection in<br />
iris scanning is perhaps more important than ever. A coalition of universities from around the world has<br />
once again convened to test liveness detection in iris scanning technologies, and is inviting academic and<br />
corporate organizations to take part in the competition. Liveness Detection Competition Arrives at Critical<br />
Time for Iris ScanningCalled LivDet-Iris: Liveness Detection-Iris Competition 2017, it has been organized by<br />
Clarkson University, Notre Dame University, West Virginia University, and Warsaw University of Technology.<br />
It?s a sort of sequel event to LivDet-Iris 2015, and arrives at a time when iris scanning is emerging as a major<br />
modality. LivDet-Iris 2017 is taking place as part of the International Joint Conference on Biometrics (ICJB<br />
2017), which will run from October 1st to 4th in Denver, Colorado.<br />
Read more: http://iris2017.livdet.org<br />
Mult ifact or aut hent icat ion sees 40% growt h in 2016<br />
The number of firms using multifactor authentication either for employees or customers leaped this year,<br />
according to a new research report. A study by SecureAuth Corp has found that multi-factor authentication<br />
jumped by more than 40% year-over-year in 2016. Of the organisations surveyed, more than half (51%)<br />
were using MFA across the organisation, while 38% have implemented it in some areas. Meanwhile, next<br />
year more than 30% of organizations are looking to expand or implement MFA in the next 12 months.<br />
?Using a second-factor can be a deterrent but is no longer enough against attacks, and organizations must<br />
evolve their methods to safeguard critical points of access such as Single Sign-On (SSO) portals and the<br />
VPN,? said Keith Graham, CTO of SecureAuth. ?By implementing adaptive access authentication,<br />
organizations can both eliminate that threat vector and provide an outstanding user experience. The latest<br />
advances in adaptive authentication include transparent techniques, such as device recognition,<br />
geo-location, the use of threat services, and even behavioral biometrics.?<br />
Read more: http://www.planetbiometrics.com/article-details/i/5354/
Biometrics I nterview<br />
Quest ion: US-VISIT was a<br />
program started after 9/11. How<br />
did US-VISIT evolve to OBIM<br />
today? What is the biggest impact<br />
that OBIM has had on U.S.<br />
national security so far?<br />
Kennet h: The US-VISIT Program<br />
was created in 2003 to develop<br />
an entry/exit system for the<br />
United States, and to serve as a<br />
biometric service provider to the<br />
immigration components of DHS,<br />
also created in 2003, through the<br />
operation of the Automated<br />
Biometric Identification System<br />
(IDENT), the central DHS<br />
repository for matching, storing,<br />
and sharing biometric and<br />
associated biographic<br />
information. Over time, as the<br />
use of biometric identification<br />
was expanded, US-VISIT?s mission<br />
grew beyond its original<br />
mandate.<br />
In March 2013, Congress<br />
recognized OBIM as the<br />
Department?s designated<br />
provider of enterprise level<br />
biometric identity services to<br />
customers within DHS, other<br />
stakeholders in the U.S.<br />
Government, and partner<br />
nations. OBIM is different than<br />
US-VISIT in that we now operate<br />
with a clear view of our role as a<br />
service provider supporting<br />
stakeholders on the front lines of<br />
homeland security. There are<br />
some legacy US-VISIT functions<br />
that no longer reside within<br />
OBIM, such as biographic<br />
entry/exit and overstay analysis.<br />
These functions were transferred<br />
to U.S. Customs and Border<br />
Protection (CBP) and U.S.<br />
Immigration and Customs<br />
Enforcement (ICE), respectively.<br />
The biggest impact that OBIM<br />
has had on U.S. national security<br />
stems from improvements in<br />
interoperability and information<br />
sharing that directly support four<br />
key Homeland Security Missions:<br />
Prevent Terrorism and Enhance<br />
Securit y; Secure and Manage<br />
Our Borders; Enforce and<br />
Administ er Our Immigrat ion<br />
Laws; and St rengt hen Nat ional<br />
Preparedness and Resilience.<br />
IDENT now stores approximately<br />
204.7 million unique identities. It<br />
processes on average more than<br />
280,000 biometric transactions a<br />
day, more than 130,000 of which<br />
come from CBP as daily U.S. entry<br />
transactions. Automated<br />
interoperability between IDENT<br />
and the Federal Bureau of<br />
Investigation?s (FBI) Next<br />
Generation Identification (NGI)<br />
biometric repository amplifies the<br />
effectiveness of the two systems.<br />
OBIM also conducts searches and<br />
transactions for the U.S.<br />
Department of State, and<br />
continues to work toward<br />
automated interoperability with<br />
the U.S. Department of Defense?s<br />
Automated Biometric<br />
Identification System (ABIS). In<br />
September 2014, OBIM and the<br />
FBI?s Terrorist Screening Center<br />
(TSC) established a direct system<br />
link between IDENT and the TSC?s<br />
Terrorist Screening Database,<br />
enabling faster, more efficient<br />
sharing of biometric information<br />
for the identification of known or<br />
suspected terrorists (KST). Finally,<br />
OBIM shares information with<br />
Australia, Canada, New Zealand,<br />
and the United Kingdom through<br />
the Five Country Conference.<br />
These collaborations vastly<br />
increase the value of OBIM?s<br />
biometric identity services.<br />
Quest ion: What effort does<br />
OBIM make to ensure privacy of<br />
those individuals who get<br />
enrolled in IDENT?<br />
Kennet h: Privacy plays an<br />
integral part of OBIM?s biometric<br />
services from conception through<br />
the planning, development, and<br />
execution of its systems and<br />
processes. The organization has a<br />
dedicated privacy team to ensure<br />
that privacy protections are built<br />
into its programs and systems to<br />
make the safeguarding of<br />
personal information a priority.<br />
Personal information is used only<br />
for the purposes for which it is<br />
collected, unless specifically<br />
authorized or mandated by law.<br />
OBIM monitors its systems and<br />
has security practices in place to<br />
protect the privacy of those<br />
whose data are collected, and to<br />
ensure the integrity of that data.<br />
Kennet h D. Gant t is the<br />
Deputy Director of the Office<br />
of Biometric Identity<br />
Management (OBIM) within<br />
the National Protection and<br />
Programs Directorate (NPPD<br />
of the Department of<br />
Homeland Security (DHS).<br />
OBIM is the lead entity for<br />
biometric identity<br />
management services. Mr.<br />
Gantt?s responsibilities<br />
include working with OBIM?s<br />
customers, stakeholders, and<br />
DHS mission partners to<br />
expand the program?s<br />
services to ensure a safe,<br />
secure, and resilient<br />
homeland.
Biometrics I nterview<br />
All sharing of personal<br />
information is authorized by law<br />
and governed, as appropriate, by<br />
information sharing agreements<br />
that ensure compliance with<br />
privacy requirements. OBIM also<br />
conducts privacy impact<br />
assessments, which provide a<br />
transparent view of the personal<br />
information maintained and how<br />
that information is used, shared,<br />
and stored.<br />
All OBIM personnel and users of<br />
the program?s data are<br />
responsible and accountable for<br />
treating personally identifiable<br />
information in accordance with<br />
the Fair Information Practice<br />
Principles (FIPPS).<br />
Quest ion: Could you list three<br />
desired advancements in the<br />
biometrics technologies that<br />
would allow for OBIM?s biometric<br />
identification services to operate<br />
more efficiently and effectively?<br />
Kennet h: OBIM works each day<br />
with its customers and other<br />
stakeholders to pursue faster,<br />
more efficient, cost effective, and<br />
accurate biometric identity<br />
services. Three technological<br />
advancements in this realm<br />
would include the ability to utilize<br />
multiple hardware and software<br />
solutions; offer advanced<br />
reporting; and make use of<br />
different storage and retention<br />
needs, as we don?t need all of our<br />
records in an active mode. More<br />
generally, a fourth advancement<br />
would involve greater use of<br />
multimodal biometrics, including<br />
iris and face. While IDENT<br />
continues to perform well, it is<br />
more than two decades old, and<br />
OBIM is planning for its<br />
replacement with a new and<br />
more robust system, the<br />
Homeland Advanced Recognition<br />
Technology (HART). HART<br />
promises a more flexible and<br />
scalable architecture that will<br />
more efficiently support core DHS<br />
missions and operations with<br />
multimodal capabilities, more<br />
efficient system processing,<br />
increased capacity and storage,<br />
and web portal access and web<br />
services.<br />
A good example of a recent<br />
technological advancement that<br />
had a significant impact on<br />
national security was as follows:<br />
"At U.S. air port s of ent ry, t he<br />
t ime needed for CBP t o send a<br />
request on an individual t o<br />
IDENT and NGI and receive a<br />
response used t o t ake bet ween<br />
24 and 72 hours, meaning t hat<br />
t he individual had left t he<br />
airport by t he t ime t he<br />
response arrived. A program<br />
called Rapid Response<br />
compressed t his t urnaround<br />
t ime t o 15 seconds, allowing<br />
CBP t o det ermine t he<br />
admissibilit y of individuals in<br />
real t ime. Over t he last 12<br />
mont hs, IDENT has processed<br />
1.42 million Rapid Response<br />
searches. "<br />
Quest ion: One of the major<br />
vulnerabilities of a biometric<br />
system is presentation of a spoof<br />
or an altered biometrics sample.<br />
What is OBIM?s position on this<br />
issue?<br />
Kennet h: OBIM watches for<br />
those individuals who try to<br />
impede law enforcement or<br />
obtain unwarranted privileges<br />
and benefits by spoofing<br />
biometric systems. While IDENT<br />
returns an automated answer for<br />
the vast majority of biometric<br />
queries, about 2,000 sets of<br />
fingerprints a day must be<br />
resolved by fingerprint examiners<br />
in OBIM?s Biometric Support<br />
Center (BSC). These highly trained<br />
professionals provide OBIM?s<br />
best defense against spoofing.<br />
Analyzing prints on<br />
high-definition work stations, BSC<br />
experts can identify the markers<br />
of altered biometrics, as well as<br />
areas that have not been altered.<br />
The BSC furnishes assured<br />
biometric identification services<br />
24 hours a day, including 10-print<br />
comparisons and verification,<br />
unknown deceased<br />
identifications, latent print<br />
comparisons, and enrollment of<br />
biometric records from law<br />
enforcement and intelligence<br />
agencies.<br />
CALL FOR ARTICLES<br />
The editor ial boar d of IEEE<br />
Biometr ics Council New sletter<br />
invites shor t ar ticles r elated to<br />
- Recent new s r elated to<br />
biometr ics<br />
- Job-posting including<br />
PhD/PostDoc positions,<br />
- Biometr ics Database<br />
- Open Sour ce Code<br />
Please send the shor t ar ticles to:<br />
hugomcp@di.ubi.pt
Spotlight<br />
In this spotlight we would like to point out the relationship between two cool research topics, namely biometric<br />
recognition and deep learning, and between them and industry latest trends.<br />
It is quite usual that successful research products reach the industry application after that consolidated results<br />
testified their effectiveness. Biometrics and deep learning are not an exception. In fact, leaving out fingerprint<br />
recognition that has a long story of application in both forensics and security, the other traits have a much more<br />
recent presence in real-world applications. The first mention of face recognition out of the fields of experimental<br />
psychology and cognitive neurosciences dates back to ?60, but entails a semi-automatic (forensic) process supported<br />
by human experts. In 1988, Kirby and Sirovich applied principle component analysis to the face recognition problem<br />
and following this work, in 1991, Turk and Pentland proposed Eigenfaces. However, the approach was successful with<br />
a small dataset and in extremely controlled conditions. In order to have the first commercial off-the-shelf (COTS) face<br />
recognition systems, though still limited to well controlled scenarios, we have to wait until 1997. Joseph Atick, a<br />
pioneer of facial recognition, helped developing the first facial recognition system in the world and founded the<br />
American company Visionics, one of the first companies in the field, in 1994. In August 1997, Visionics announced the<br />
first large scale shipping of its FaceIt(R) face recognition software engine to application developers and system<br />
integrators. The engine was encapsulated into a powerful Software Developer Kit to enable developers to quickly<br />
incorporate facial recognition capabilities into their applications, even beyond the forensics context. Afterwards, in<br />
November 1998, Visionics Corporation announced that its FaceIt(R) face recognition software was the first software of<br />
its kind to be used in a Closed Circuit Television (CCTV) control room application. The new CCTV-based anti-crime<br />
system called 'Mandrake' was installed in the borough of Newham, East London, in the United Kingdom. In March<br />
2000 two early direct competitors, Viisage Technology, Inc. and Biometrica Systems, Inc. announced the fiftieth order<br />
and installation of its full surveillance suite of Visual Casino loss-reduction systems, at Mirage Resort, Las Vegas.<br />
Visual Casino, a computer application with its facial recognition technology, was used by surveillance operators to<br />
identify cheaters and other casino undesirables, as well as casino VIPs. After that new competitors entered the<br />
market, being Cognitec one of e most renowned nowadays.<br />
Another biometric trait that has reached the market is iris, with a similar history from the first significant work<br />
presented by J. Daugman in 1993, except for the fact that it often requires special equipment. . In 1995, the first<br />
commercial product was distributed on the market by Iriscan, Inc., that become Iridian Technologies, Inc. in 2000<br />
(that in 2006 was acquired by Viisage Technologies, Inc.). In 2005, the broad patent covering iris recognition expired,<br />
so that other companies could develop their own algorithms for iris recognition. The patent on the IrisCodes® has<br />
expired too in 2011. Going to deep learning, it gave a second life to neural networks, and is gaining increasing<br />
attention in industry too. Notwithstanding its definition as a ?a new area of Machine Learning research?<br />
(http://deeplearning.net) many researchers consider the Neocognitron introduced by Kunihiko Fukushima in 1980 as<br />
one of the first ancestors of deep learning architectures, in particular of those deriving from artificial neural networks<br />
(ANN). However, it is only after 2007 that the work by Yoshua Bengio renewed the research community interest on<br />
related topics. ?The large-scale impact of Deep Learning in Big Tech Companies began around 2010 with speech<br />
recognition. ? It took around 30 years to become mainstream. Computers were not powerful enough and companies<br />
didn?t have such a large amount of data.? (http://tinyurl.com/lnv77uo). At present, Skymind offers to enterprise deep<br />
learning for big data, e.g., company fraud detection in finance, recommender systems in e-commerce, and anomaly<br />
detection for hardware makers. But, as noticed in December 2014 by Forbes (http://tinyurl.com/kwswv7p) big<br />
American companies like Google, IBM, Microsoft, Facebook and China?s Baidu are well-represented in applications<br />
providing the core technologies for applications addressing a radical change in enterprise and industry.<br />
Last but not least, the use of deep learning as a strategy for biometric extends from research to industry and to social<br />
networks. DeepIris by CASIA (Chinese Academy of Science) is one of the latest proposals for iris recognition. But in the<br />
?real world? Google already uses deep learning for voice recognition on Android phones, and Facebook uses DeepFace<br />
to identify friends in users?photographs. To this aim, the latter exploits ?a face representation from a nine-layer deep<br />
neural network. This deep network involves more than 120 million parameters using several locally connected layers<br />
without weight sharing, rather than the standard convolutional layers. Thus we trained it on the largest facial dataset<br />
to-date, an identity labeled dataset of four million facial images belonging to more than 4,000 identities.?<br />
(http://tinyurl.com/kco8e7q)<br />
What is the key for this successful marriage? According to many observers, the availability of big amounts of data is a<br />
possible explanation of recent achievements. From one side, it might represent a possible stressing factor for<br />
machine learning methods. However, the increased computing power and storage easily compensate this. On the<br />
other side, the main the positive aspect of this wealth of data is the possibility to carry out a huge training and a<br />
fine-tuning of a large number parameters relevant to classification, that was impossible before.
Paper Summary<br />
Overview:<br />
Cardiovascular Biom et rics: Com bining Mechanical and Elect rical Signals (* )<br />
Ikenna Odinaka a , Joseph A. O'Sullivan b , Erik J. Sirevaag c , John W. Rohrbaugh c<br />
ikenna.odinaka@duke.edu<br />
a - Duke University, b- Washington University in Saint Louis<br />
c - Washington University School of Medicine<br />
Two biometric traits that have been investigated for their potential use in human recognition are the<br />
electrocardiogram (ECG) and the laser Doppler vibrometry (LDV) signal. Although the ECG and LDV signals are both<br />
measures of cardiovascular activity, the differences between them are substantial and important, and reflect<br />
fundamentally different aspects of cardiovascular performance. Whereas the ECG is an electrical signal, which is<br />
generated in the myocardium and is conducted by volume to the surface electrodes, the LDV signal (recorded here<br />
from the neck, overlying the carotid artery) derives from mechanical displacements of the skin overlying the carotid<br />
artery. Previous studies have shown that biometric systems based on the ECG or LDV signal can achieve single digit<br />
authentication performance, in terms of equal error rate (EER). In particular, EERs between 5 and 7% have been<br />
reported, when training and testing data are obtained from different sessions [1], [2]. In this work, we propose<br />
combining ECG and LDV signals for human recognition. In addition, we employ standard fusion techniques for<br />
aggregating the information provided by each biometric system, to achieve a substantial increase in recognition<br />
performance. We show that a reduction by at least half in the EER can be obtained from this fusion. Moreover, we<br />
report on the methods we have found to work best for authentication and identification. The principal goal of this<br />
effort is thus to assess authentication and identification performance using combined ECG and LDV signals. We<br />
pursued this objective using a number of well-established and accepted fusion methods.<br />
A multibiometric system is one that uses more than one biometric trait (multimodal), biometric classifier<br />
(multi-algorithm), instance of the same biometric trait (multi-instance), sensor for extracting the same biometric trait<br />
(multi-sensor), or sample of the same biometric trait (multi-sample) for biometric recognition [3]. A multi-instance<br />
system applies to biometric traits that do not occur singly; e.g., ten fingers, two irises, and multiple ECG channels. A<br />
multi-sample system combines multiple samples or different portions of the same biometric trait; e.g. the front and<br />
side of a face, segments of the same finger, multiple ECG heartbeats. For physiologically related signals, which change<br />
over time, we can combine signals from more than one session, the so called multi-session system [1]. In this work,<br />
we present the performance gains that result from a multimodal, multi-session, multi-sample system involving the<br />
ECG and LDV signals. The individual systems that constitute a multibiometric system can be combined at multiple<br />
levels, ranging from the raw signal to decision processes [4]. In this work, we compare the authentication and<br />
identification performance of the cardiovascular multibiometric system, when fusion is performed at the feature,<br />
score, rank, and decision levels of the biometric pipeline. In addition, we analyze the performance gains of the<br />
cardiovascular multibiometric system over the individual ECG and LDV biometric systems. The subtle differences<br />
between ECG and LDV signals may not be readily appreciated, and it may not be directly obvious that combining the<br />
ECG and LDV signals will lead to improved performance. For example, based on Figure 1, the use of temporal fiducial<br />
features may decrease the performance of the joint biometric system, since the timing information of the two signals<br />
are highly correlated. However, the waveform features of the two signals are very different, suggesting that<br />
amplitude-based, angle-based, area-based fiducial features, and non-fiducial features are better able to capture<br />
uncorrelated information from the ECG and LDV signals. We adopted the short-time Fourier transform approach to<br />
feature extraction. This is by no means the only way the features can be gotten from each modality for later fusion.<br />
See the paper by Odinaka et al. [2] for a review of the feature extraction methods that have been proposed for ECG<br />
biometrics.<br />
Informat ion Fusion:<br />
1) Feature-level Fusion: We concatenate the features vectors from the different modalities to obtain a fused vector. A<br />
classifier is then employed on the fused vectors.<br />
2) Score-level Fusion: We explored the three major approaches to score-level fusion: score normalization, density<br />
estimation, and classifier-based fusion [3], [4]. Examples of score normalization techniques include: Min-Max, Z-Score<br />
(Mean-Std), Median-MAD (median absolute deviation).<br />
(* ) Appeared in IEEE Transactions on Information Forensics and Security, 10(1):16-27. January, 2015.
Paper Summary<br />
3) Rank-level Fusion: After computing the match scores of each test signal, each biometric matcher (ECG or LDV)<br />
outputs a rank for each of the classes (enrolled individuals). We used techniques including the highest rank, Borda<br />
count, and the weighted Borda count (logistic regression) [5], to obtain the fused rank.<br />
4) Decision-level Fusion: The decision made by a biometric matcher can be an accept or reject, in authentication<br />
mode, or an identity or reject, in identification mode. Example techniques we utilized to fuse the decisions from both<br />
biometric systems, include, ?AND? and ?OR? rules [6].<br />
Fig. 1: A 5s segment of a subject?s simultaneously recorded ECG and LDV velocity waveforms.<br />
Result s:<br />
Using the score-level fusion techniques gave the best multibiometric performance improvement for both<br />
authentication and identification modes of operation. The authentication performance, as measured by EERs,<br />
improved from ~6% for each of the unimodal ECG and LDV biometric systems, to ~2% for the multibiometric system.<br />
Moreover, in the identification mode, the rank-1 accuracy improves from ~80% for each unimodal biometric system,<br />
to ~92% for the multibiometric system.<br />
Conclusions:<br />
We performed a large scale cardiovascular biometric experiment on 258 individuals from three sessions. This<br />
represents a significant extension of the results of uni-variate modeling [1], [2] to a multidimensional case. We<br />
performed a comprehensive study of the performance of the multibiometric system at different levels of the<br />
biometric pipeline, using standard fusion techniques. In one measure of performance, namely EERs, we showed that<br />
performance improves from about 5.3% to about 2.1%. Based on our analysis, we made recommendations on the<br />
methods that we found to work best in the authentication and identification modes of operation.<br />
References:<br />
[1] M. Chen, J. A. O?Sullivan, N. Singla, E. J. Sirevaag, S. D. Kristjansson, P.-H. Lai, A. D. Kaplan, and J. W. Rohrbaugh,<br />
?Laser Doppler vibrometry measures of physiological function: evaluation of biometric capabilities,? TIFS, vol. 5, no. 3,<br />
pp. 449?460, 2010.<br />
[2] I. Odinaka, P.-H. Lai, A. D. Kaplan, J. A. O?Sullivan, E. J. Sirevaag, and J. W. Rohrbaugh, ?ECG biometric recognition: A<br />
comparative analysis,? TIFS, vol. 7, no. 6, pp. 1812?1824, Dec. 2012. [3] A. K. Jain and A. Ross, ?Multibiometric systems,?<br />
Commun. ACM, vol. 47, pp. 34?40, Jan. 2004.<br />
[4] A. Ross, K. Nandakumar, and A. K. Jain, Handbook of Multibiometrics. Springer-Verlag New York, Inc., 2006.<br />
[5] T. K. Ho, J. Hull, and S. Srihari, ?Decision combination in multiple classifier systems,? PAMI, vol. 16, no. 1, pp. 66?75,<br />
Jan 1994.<br />
[6] J. Daugman, ?Combining Multiple Biometrics.? http://www.cl.cam.ac.uk/_jgd1000/combine/combine.html, 2000.
Conference Report - BI OSI G 2016<br />
The 15th edition of the International Conference of<br />
the Biometrics Special Interest Group (BIOSIG) took<br />
place at Fraunhofer IGD in Darmstadt, Germany<br />
from September 21 to 23 and attracted registered<br />
more than 100 participants. This year participants<br />
even travelled all the way from Japan, Korea,<br />
Argentina, Uruquay and the US to Darmstadt in<br />
order to join the BIOSIG community.<br />
The program was composed of scientific research<br />
contributions on the one hand and reports about<br />
large-scale applications on the other hand. The<br />
opening keynote talk was given by Richard Rinkens<br />
(EC, DG Home), who presented the new proposal<br />
for the Entry Exit System (EES), adopted by the<br />
European Commission after an intense period of<br />
analysis and real-world testing (eu-LISA has tested<br />
13 entries - sea, rail, air - with different options).<br />
The new proposal is based on multi-modal<br />
biometrics combining 4-fingerprints and a good<br />
quality facial-image taken live. In particular, new<br />
procedures for EU citizens, 3rd country nationals<br />
with visa and 3rd country without visa are<br />
considered. One of the main objectives is the<br />
identification of overstayers and to enable<br />
subsequent evidence based visa policy making. To<br />
tackle these issues, the EES will substitute manual<br />
procedures with electronic stamps in an electronic<br />
database. In addition, a new link has been made<br />
between EES and VIS, and a common AFIS should<br />
be deployed for the three systems. Future features<br />
include a webservice to check credentials and<br />
authorized time left, which will be the first<br />
consultable service for an EU system of this kind<br />
through internet. Finally, it was highlighted that the<br />
EES will seriously reinforce internal security and<br />
fight against terrorism.<br />
Accepted conference contributions included 16<br />
presentations and covering soft biometrics for face<br />
or iris, fingerprint, vein, face, iris, voice, gait and<br />
keystroke recognition. Other challenges addressed<br />
included facial forensics use cases, voice activity<br />
segmentation, 3D gloves for contactless verification,<br />
new biometric template protection schemes or the<br />
generation of synthetic fingerprint alteration<br />
database to allow the further development of<br />
algorithms to detect them. Another relevant topic<br />
covered in the conference was Presentation Attack<br />
Detection (PAD) for characteristics such as face and<br />
iris. The poster session with 25 contributions was a<br />
good mix of research results from academic and<br />
industrial research labs and visitors did spend a<br />
long time in the poster exhibition before the start<br />
of the social event with the traditional late summer<br />
barbeque ? providing lots of opportunities for<br />
networking.<br />
On the last day of the conference, Arun Ross<br />
addressed the issue of personal privacy in<br />
biometric systems. In particular, he considered<br />
several questions such as (a) Can additional<br />
information about an individual be automatically<br />
gleaned from biometric data? (b) Can biometrics be<br />
used to surreptitiously track an individual? (c) Can<br />
multiple biometric databases be linked to develop a<br />
more complete profile of an individual? (d) Who<br />
owns the biometric data collected from an<br />
individual, and how long should the biometric data<br />
be retained in an identity management system? He<br />
started the keynote with a definition stemming<br />
from 1890: ?privacy is the right to be left alone?.<br />
However, in contemporary society it is easy for<br />
someone to take a picture of you and find out who<br />
you are by harnessing the power of automated face<br />
recognition and cloud computing. In the first part<br />
of the talk, he discussed methods by which<br />
biographical (e.g., gender from fingerprints),<br />
anatomical (e.g., crypts in iris), sensorial (e.g., type<br />
of device used to image a certain fingerprint), and<br />
environmental (e.g., intensity of ambient lighting<br />
during iris image acquisition) information can be<br />
gleaned from biometric data.
Conference Report - BI OSI G 2016<br />
While the resultant information can be used to<br />
improve person recognition performance in the<br />
framework of ?soft? biometrics or forensics, the<br />
extraction of such information can be deemed to<br />
violate privacy. He gave the example where face or<br />
iris data can be surreptitiously used to divulge<br />
pathological information about an individual.<br />
In the second part of the talk, he discussed ways by<br />
which privacy can be accorded to stored biometric<br />
data. The proposed methods rely on the principles<br />
of visual cryptography and signal mixing to<br />
generate biometric templates that can suppress<br />
some of the additional information about a person<br />
that is resident in the biometric data. He reported<br />
experimental results discussing the efficacy of<br />
these methods. The talk concluded by pointing out<br />
that privacy enhancing technologies can be<br />
judiciously used by biometric systems to ensure<br />
that the benefits of biometrics are not undermined<br />
by privacy concerns.<br />
The last conference day concluded with a keynote<br />
by Davide Maltoni addressing the issue of<br />
large-scale verification within projects such as<br />
UIDAI, BMS, ePassports or SIS-II: predicting<br />
accuracy of biometric systems is a very difficult<br />
problem. Statistical modelling techniques often<br />
require to make assumptions on data distributions<br />
that we cannot validate in practice. A different<br />
approach is running fingerprint recognition<br />
algorithms on large datasets of synthetic data.<br />
However, we have to face two problems: i) ensure<br />
that synthetic data well approximate real data; ii)<br />
running huge amount of fingerprint comparisons in<br />
a reasonable time. In particular, the SFinGe<br />
software developed at the University of Bologna<br />
within the EU FIDELITY project can mitigate this lack<br />
of large databases, where not only one but several<br />
samples of a given synthetic fingerprint can be<br />
generated modelling distortion. For the empirical<br />
evaluation and comparison with real data,<br />
verification was run using Minutiae Cilynder Codes<br />
(MCC), which model spatial and directional<br />
contributions. The software has been optimized for<br />
GPUs, needing only 5.6 ms for a single identification<br />
on 250K subjects (44.6 millions of comparisons per<br />
seconds). The empirical evaluation on synthetic data<br />
yielded predictions similar to those found in a real<br />
case study: IUDAI. As future actions an analysis with<br />
NFIQ2.0 will be considered and scaling up to a<br />
database modelling the world population (7 billion),<br />
which will need about 3 months for generation and<br />
a few weeks for identification.<br />
As in previous BIOSIG conferences participants of<br />
the conference themselves voted for the best paper<br />
and the best poster that was presented at the<br />
conference. The winner of the BIOSIG 2016 best<br />
paper award is Sunpreet S. Arora (Michigan State<br />
University) for his presentation ?3D Whole Hand<br />
Target: Evaluating Slap and Contactless Fingerprint<br />
Readers?, which convinced the majority of the<br />
participants.<br />
While the poster session showed many impressive<br />
research results that stimulated long discussions,<br />
there was one contribution, which was chosen by<br />
the participants as best and it received thus the<br />
BIOSIG 2016 best poster award. It was the poster of<br />
Yoshinori Koda (NEC Corporation) with the title:<br />
?Advances in Capturing Child Fingerprints: A High<br />
Resolution CMOS Image Sensor with SLDR Method?.<br />
The BIOSIG conference was preceded by the 3rd<br />
EAB Research Project Conference and was further<br />
co-located with two satellite workshops:<br />
The meeting of the TeleTrusT Biometric Working<br />
Group and joint meeting of the ethical committee of<br />
the European Association of Biometrics.<br />
The 2016 BIOSIG conference was jointly organized<br />
by the Competence Center for Applied Security<br />
Technology (CAST) and the special interest group<br />
BIOSIG of the Gesellschaft für Informatik e.V. (GI).<br />
The conference was technically co-sponsored by<br />
IEEE Biometric Council and the papers will be added<br />
to IEEE Xplore.
Calender<br />
Call for Proposal for IEEE Conference on Biomet rics Theory, Applicat ion<br />
and Syst ems (BTAS), 2018<br />
Deadline: May 31, 2017<br />
BTAS is the flagship conference for IEEE Biometrics Council. Each proposal<br />
should include information on (but not limited to):<br />
- Organizing and technical committees, at least general and program<br />
chairs<br />
- Verified venue and dates (without possible conflict with other major<br />
conferences in closely related fields). Please provide venue contact info.<br />
The venue is not restricted to be in Washington, DC<br />
- Verified accommodation arrangements;<br />
- Publicity and plans for growing the conference.<br />
- Timetable with critical milestones;<br />
- Budget estimates using the standard budget template available at:<br />
http://www.ieee.org/documents/financial_reporting_tool.xls<br />
- Possible external sponsors.<br />
Further guidelines on IEEE conference organization can be found at<br />
http://www.ieee.org/conferences_events/conferences/organizers/index.html<br />
Please send proposals to Sebastien Marcel, Vice-President for Conferences, IEEE<br />
Biometrics Council at marcel@idiap.ch<br />
The proposals will be vetted by the IEEE Biometrics Council Conference<br />
Committee and result announced through IEEE Biometrics mailings and the<br />
website.<br />
May 30 - Jun 3, 2017<br />
IEEE FG2017<br />
http://w w w.fg2017.or g/<br />
Sep 20 - 22, 2017<br />
BIOSIG 2017<br />
www.biosig.de/biosig2017<br />
Oct 1 - 4, 2017<br />
IEEE/IAPR IJCB2017<br />
http://w w w.ijcb2017.or g/<br />
16t h Int ernat ional Conference of t he<br />
Biomet rics Special Int erest Group<br />
(BIOSIG 2017)<br />
Sep 20 ? 22, 2017<br />
Darmstadt, Germany<br />
www.biosig.de/biosig2017<br />
Important Dates:<br />
- Full Paper Submission deadline: 30 May<br />
- Notification of acceptance date: 15 Jul<br />
- Final submission deadline: 15 Aug<br />
2017 IEEE/IAPR Inter nat i onal Joi nt<br />
Confer ence on Bi om et r i cs (IJCB 17)<br />
The Inter national Joint Confer ence on<br />
Biometr ics (IJCB 2017) combines tw o major<br />
biometr ics r esear ch annual confer ences, the<br />
Biometr ics Theor y, Applications and Systems<br />
(BTAS) confer ence and the Inter national<br />
Confer ence on Biometr ics (ICB). IJCB 17 w ill be<br />
held fr om October 1 till October 4, 2017 in<br />
Denver , CO. IJCB 2017 solicits submissions<br />
r elated to biometr ics. Call of paper s is available<br />
at: http://w w w.ijcb2017.or g/<br />
Paper submission deadline is Apr il 15, 2017.<br />
Paper submission instr uctions:<br />
http://w w w.ijcb2017.or g/ijcb2017/author s.php<br />
Paper submission site:<br />
https://cmt3.r esear ch.micr osoft.com/IJCB2017/<br />
SI BWild 2017: Image and Vision Comput ing (IVC) Special Issue<br />
on Biomet rics in t he Wild<br />
Bir Bhanu, Abdenour Hadid, Qiang Ji, Mark Nixon, Vitomir Struc<br />
Paper Submission Deadline: Jun 30, 2017<br />
Publication Date: April 2018<br />
http://tinyurl.com/kp5dxxg<br />
SI VSB2017: Pat t ern Recognit ion Let t ers (PRL) Special Issue on<br />
Video Surveillance-orient ed Biomet rics<br />
Changxing Ding, Kaiqi Huang, Vishal M. Patel, Brian C. Lovell<br />
Paper Submission Deadline: March 30, 2017<br />
http://tinyurl.com/l7mhhwr<br />
SI UER 2017: IET Biomet rics Special issue on Unconst rained Ear<br />
Recognit ion<br />
Peter Peer, Vitomir Struc<br />
Paper Submission Deadline: Sep 1, 2017<br />
Publication Date: May, 2018<br />
http://tinyurl.com/n8svouv