International Journal of Computer Information Systems and Industrial Management Applications
ISSN 2150-7988 Volume 3 (2011) pp. 174-184
© MIR Labs, www.mirlabs.net/ijcisim/index.html
Dynamic Publishers, Inc., USA
Security of Biometric Authentication Systems
Vashek Matyáš 1
, Zdeněk Říha 2
1
Faculty of Informatics, Masaryk University
matyas@fi.muni.cz
2
Faculty of Informatics, Masaryk University
zriha@fi.muni.cz
Abstract: This paper outlines our views of actual security of
biometric authentication and encryption systems. The
attractiveness of some novel approaches like cryptographic key
generation from biometric data is in some respect
understandable, yet so far has lead to various shortcuts and
compromises on security.
This paper starts with an introductory section that is followed
by a section about variability of biometric characteristics, with a
particular attention paid to biometrics used in large systems.
The following sections then discuss the potential for biometric
authentication systems and for the use of biometrics in support
of cryptographic applications as they are typically used in
computer systems.
Keywords: IT security, biometrics, authentication.
I. Introduction
This paper summarizes our opinions and findings after more
than a decade of studying biometric authentication systems
and their security. This is an extended version of our work
presented at the Computer Information Systems and Industrial
Management Applications (CISIM) 2010 conference, held in
Cracow, Poland. The paper summarizes our personal views
and opinions on selected issues in the area of biometric
authentication and related areas of cryptography.
Proper user identification/authentication is an essential
requirement for reliable access control and as such is a key
enabler for electronic commerce. There are three types of
identification/authentication methods that can be used either
on their own or in various combinations. Identity verification
(authentication) in computer systems has been traditionally
based on something that one has (key, magnetic or chip card)
or one knows (PIN, password). Biometrics are based on the
principle of measurable physiological or behavioural
characteristics such as a fingerprint or a voice sample.
Biometric systems can be used in two different modes.
Identity verification (also called one-to-one comparison or
authentication) occurs when the user1
claims to be already
enrolled in the system (presents an ID card or login name); in
this case the biometric data obtained from the user are
compared to the user's data already stored in the database.
1
We shall use the term ‘data subject’ according to [32] later in the paper when
we relate the user directly to his/her biometric data.
Identification (also called search, recognition or one-to-many
comparison) occurs when identity of the user is a priori
unknown. In this case the user's biometric data are compared
against all records in the database as the user can be anywhere
in the database or she actually does not have to be there at all.
Authentication is typically a pre-requirement of authorization
(to log in, to access files, to enter an aircraft, etc.). While
biometric authentication is attractive because it principally
authenticates the user (and not something that can be
disclosed or passed to a colleague), its shortcomings relate to
problems with accuracy, privacy protection, secrecy of the
biometric data and therefore to the need for a reliable liveness
testing.
Before a user can be successfully verified or identified by the
system, she must be registered with the biometric system.
User's biometric data are captured, processed and stored. The
process of the user's registration with the biometric system is
called enrolment.
There are basically two kinds of biometric systems:
 Automated identification systems operated by
professionals (e.g., police Automated Fingerprint
Identification Systems – AFIS). The purpose of such
systems is to identify an individual in question or to find
an offender of a crime according to trails left at the crime
scene. Enrolled users do not typically have any access to
such systems and operators of such systems do not have
many reasons to cheat.
 Biometric authentication systems used for access control.
These systems are used by ordinary users to gain a
privilege or an access right. Securing such a system is a
much more complicated task.
It is worth noting that the involvement of a human factor in
the former type of biometric system enormously reduces the
problems of the latter type of system. This paper will focus on
the latter type of systems as improvement of their security,
without human intervention in the process, has considerably
more impact on computer security and also is more
challenging.
Matyáš and Říha175
II. Accuracy of biometric systems
The most significant difference between biometric and
traditional authentication techniques lies in the answer of the
biometric system to the authentication/identification request.
Biometric algorithms do not give simple YES/NO answers.
Instead, we are being told how similar or dissimilar the
current biometric data are to the record stored in the database.
We have to allow for some variability of the biometric data in
order not to reject too many authorized users (this would be a
case of the false rejection error). However, the greater
variability we allow the greater is the probability that an
impostor with similar biometric data will be accepted as an
authorized user (this is a case of the false acceptance error).
The variability allowed is usually called a (security) threshold
or a (security) level.
A. Variability of biometric characteristics
The performance of a biometric technique depends on which
characteristics – whether genotypic or phenotypic – the
technique is based on. Genotypic characteristics do not
change over time. This is good news for the false rejection
rate which may remain low as the matching algorithm does
not have to adapt to changes. The bad news is that genotypic
characteristics cannot distinguish monozygotic twins. So the
percentage of identical twins2
in population sets the lower
limit on the false acceptance rate (so called genotypic error
rate). The phenotypic characteristics do not set limits on the
false acceptance rate, but it is clear that the phenotypic
variation over time imposes a lower limit on the false rejection
rate (so called phenotypic error rate).
More precisely, the performance of biometric techniques is
determined by two kinds of variability among the acquired
biometric characteristics:
 Within-subject variability: As the results of biometric
measurements are never the same the biometric system
must accept similar samples stemming from one biometric
characteristic as a true match. Although the matching
algorithm may allow for a variability of the input
measurements, it is clear that higher within-subject
variability implies more false rejects. Therefore
within-subject variability sets the lower limit on the false
rejection rate.
 Between-subject variability: If between-subject
variability is low then it is more difficult to distinguish two
subjects and a false accept may occur. The lower the
between-subject variability the higher the false acceptance
rate. Therefore between-subject variability sets the lower
limit on the false acceptance rate. An ideal biometric
characteristic impacts very high between-subject
variability.
B. Error rates
The interaction with a biometric system starts with the
enrolment, where the quality of enrolment data is very
important and significantly influences the system
performance. Often several input samples (e.g., 3 or 5) are
combined to create one biometric reference (or to verify
usability of the newly created biometric reference).
2
The probability that a person has an identical twin is estimated as 0.8% [17].
The probability of a person not being able to enrol in a
biometric system is called the Fail to Enrol rate (FTE). It is
computed as a fraction of people who could not enrol in the
system out of the complete group of people. The FTE rate
includes people without fingers (for fingerprint systems),
visually impaired people (for iris-based systems), etc.
The FTE rate is generally estimated as 1-2% for fingerprint
based systems and 1% for iris based systems. Real values of
the FTE rate are dependent on the particular scanner, the
enrolment policy and the user population.
A good illustration of numbers from the real life can be the
fingerprint enrolment of epassport applicants. In Germany [7,
51] good quality optical scanners with 500 or 1000 DPI have
been used for the enrolment. During the three month
enrolment trials 99% of the applicants could be enrolled with
fingerprints of 2 fingers and 0.9% with a fingerprint of one
finger. Only 0.1% of applicants could not be enrolled by the
fingerprint system at all. Over 80% of the enrolled images
were of the highest quality (NFIQ3
=1), while approximately
2% were of low and very low quality (NFIQ=4 or NFIQ=5).
For verification/identification attempts, the biometric input
sample is obtained and its quality is verified. If the quality
does not satisfy certain minimal quality requirements, the
acquisition process must be repeated. If all repeated
acquisitions do not yield sufficiently good samples, the person
cannot be identified/verified and such an attempt increases the
Failure To Acquire (FTA) rate. Sometimes the minimal
quality can be configured and then the stricter we are with the
quality check the better results we get during the biometric
comparison and vice versa. The FTA rate can be therefore
traded off with biometric matching error rates.
Input samples of sufficient quality are processed in the
biometric matching algorithm. The matching algorithm
compares the input sample with a biometric reference (in the
case of verification) or number of references (in the case of
identification). The result of the matching algorithm is either
correct or incorrect. If an error occurs, the resulting decision
can either incorrectly refuse an authentic person (this is
so-called false non-match – FNM) or match an impostor with
another person's biometric reference (this is so called false
match – FM). What happens next depends on the system
policy. In the case of single attempt scenario, the
verification/identification ends. In the case of, for example
three-attempt scenario, re-acquisition is possible if the person
is not being recognized (either false non-match or correct
refusal of an impostor).
The final result of an authentication/verification attempt is
either correct acceptance or correct refusal, false acceptance
or false rejection. In the case of single-attempt scenario the
FRR and FAR can be computed as:
FRR = FTE + (1 - FTE) · FTA + (1 - FTE) · (1 - FTA) · FNMR
FAR = (1 - FTE)2
· (1 - FTA) · FMR
For the purpose of FAR computations the so-called
zero-effort (also called random forgery) unauthorized
authentication attempts are taken. In this case attackers are not
actively changing their biometric characteristics (for example
in the case of dynamic signature systems they sign as usual).
3
NIST Fingerprint Image Quality – quality assessment (prediction of matcher
performance) ranging from 1 (best) to 5 (worst).
Security of Biometric Authentication Systems 176
Sometimes the minimal quality required for a successful
enrolment can be configured. It is, however, clear that the
stricter we are with the quality control at the time of enrolment
(i.e., the better quality of the biometric reference), the better
results we achieve later in verification/identification attempts
and vice versa. Therefore matching error rates can be traded
off with the enrolment quality requirements. In 2004 Atos
Origin (commissioned by the UK Passport Service) ran a
biometric trial. Facial, iris and fingerprint systems were tested
in ‘real’ conditions with 3 groups of participants: Quota
(representative sample of the population), Opportunistic
(volunteers) and Disabled (several types of disabilities). The
Quota and Disabled results can be briefly summed in the table
1. The results are not particularly good and the study points to
negative extremes from the biometric usability point of view.
For details (explanation of some of the results, shortcomings
of the trial, etc.) see the final report of the trial [8].
Face
FTE FTA FNMR FRR
Quota 0.15 0.00 30.82 30.92
Disabled 2.27 0.00 51.57 52.67
Iris
FTE FTA FNMR FRR
Quota 12.30 0.44 1.75 14.21
Disabled 39.00 0.68 8.22 44.39
Fingerprint
FTE FTA FNMR FRR
Quota 0.69 6.98 11.70 18.43
Disabled 3.91 3.14 16.35 22.14
Table 1. The error rates of facial, iris and fingerprint
systems in a UK 2004 trial [8]. All values are expressed in %.
More favourable error rates have been achieved during the
real operation of biometric systems deployed at airports. The
iris based system at Amsterdam airport [22] is operating with
FRR of 1.5%. The fingerprint based system at Paris CDG
airport [22] (when operational) was achieving similar FRR.
The biometric system at Portuguese airports [23] is based on
facial recognition and operates with FRR of 4-5%.
The correct way to calculate error rates is to compute error
rates for each person who contributes to the tests and then to
average4
the rates over the group of all the people. Otherwise
the results can be biased by an unbalanced number of
verification/identification attempts done by different people.
As we have seen, the accuracy/usability of biometric systems
can be measured in the terms of FTE, FTA, FMR, FNMR,
FAR and FRR. When comparing different systems, typically
only the resulting FR and FA rates are used. The FAR and
FRR can be graphically expressed in a FAR-FRR graph,
where both the error rates are a function of the threshold value
or can be plotted in a ROC graph where the FAR is a function
of FRR or vice-versa (thus eliminating the threshold value
from the graph). The point where FAR and FRR have the
same value is called the equal error rate (EER) or the
crossover accuracy. Such a threshold does not have a
particular importance, but the resulting EER can be used as a
4
Weighted average corresponding to the target population can also be used.
(rather simplified) performance value of a biometric system in
evaluations.
Now let us review some real numbers. There are several types
of tests [10] and not all the results must necessarily be
comparable.
The American NIST has been regularly testing the accuracy
of fingerprint, iris and facial biometric systems.
The facial recognition algorithms were tested during the
FRVT (Face Recognition Vendor Test) 2002 and FRVT 2006
tests. Iris recognition algorithms were tested in ICE (Iris
Challenge Evaluation) 2005 and 2006. Fingerprint algorithms
were tested during the FpVTE (Fingerprint Vendor
Technology Evaluation) 2003, Slapseg (Slap Fingerprint
Segmentation Evaluation) 2004 and II (2008), PFT
(Proprietary Fingerprint Template Evaluation) 2003 and II
(2010), MINEX (Minutiae Interoperability Exchange Test)
2004 and ELFTO (Evaluation of Latent Fingerprint
Technologies) 2007 tests. As an example of the results of the
NIST test effort we include here the ROC graph of facial
biometric systems from 2006. The details of the NIST tests
can be found at fingerprint.nist.gov, iris.nist.gov and
face.nist.gov.
Figure 1. The ROC graph of several facial recognition
algorithms and human ability to recognise faces (FRVT 2006
run by NIST [45] for facial images with illumination
changes).
C. Large scale systems
Designing a biometric system for few data subjects (as users
are called according to [32]) is relatively easy. Tuning a
system for millions of data subjects is significantly more
challenging.
While the verification speed and accuracy is essentially same
for a system with 10 data subjects and for a system with 10
million data subjects, the identification mode makes the
difference.
In identification mode the biometric system can incorrectly
reject the data subject (and this affects the false-negative
identification-error rate – FNIR) or incorrectly accept an
impostor (and this is measured by the false-positive
identification-error rate – FPIR).
In the case of a single attempt scenario the values of FNIR and
FPIR can be estimated from the verification FMR and FNMR
as follows:
FNIR = FTA + (1 - FTA) · FNMR
FPIR = (1 - FTA) · (1 - (1 - FMR)x
)
where x is the number of biometric references in the database.
Let us illustrate the identification accuracy with an example,
consider a biometric system with the verification FMR of
0.5% and FNMR of 5%. The false match rate of 0.5%
(FMR=0.005) may look attractive. Let us further have a small
organization with 100 data subjects. If this organization uses
the aforementioned biometric system in the identification
mode, the following identification accuracy is achieved (for
simplicity we calculate with FTA set to 0):
FPIR = 1-(1-FMR)100
= 1-0.995100
 0.3942 = 39.42%
FNIR = FNMR = 5%
A system with an FPIR of nearly 40% is useless. Even if the
verification FMR of 0.5% may look impressive it is not
suitable for identification even within a small group of dozens
of data subjects. Large scale identification systems need
algorithms with much better accuracy.
III. Secrecy and diversity of biometric data
As we all suspect, most biometric data are not secret. Photos,
fingerprints and other biometric data can be more-or-less
easily obtained by a targeted attack. Therefore security of any
system cannot be fully based on the secrecy of biometric data.
Having said that we have to admit that the need of the
acquisition of the biometric data is a factor that can make
overall attack more expensive in the terms of time, money
and/or effort. Therefore the use of biometric data as an
additional security factor can have a (security) value in some
cases. Other issues related to the secrecy of biometric data
include unchangeableness of most biometric characteristics (it
would be useful after compromise of biometric data) and
small diversity in the choice of parameters of biometric
systems (like which finger to scan) resulting in several
biometric systems storing basically the same data of a data
subject (and administrators having access to such data). Data
subjects moreover will not learn that their biometric data has
been stolen (in contrast with, e.g., tokens) until a misuse is
detected.
The secrecy of biometric data relates to ease of covert and
overt acquisition of such data and to the diversity of such data
in separate instances of biometric systems. Naturally, there
are differences between the various biometric techniques. Let
us now look at particular biometric techniques:
 Facial systems: facial biometric systems are based on 2D
or 3D representations of faces. The face is typically the
most visible part of the human body and can be easily
watched and/or photographed and/or recorded. Covert
acquisition of a 2D facial image is simple and can be made
even from a long distance (as every paparazzo could
describe). Even if faces change with time and other
influences (like fashion), people only have one face and
therefore all biometric systems based on faces use the
same information.
 Infrared facial systems use photographs or recordings of
peoples’ faces imaged in infrared wavelengths. Infrared
devices are more expensive, but are widely available. The
secrecy of such biometric data is comparable to data used
in facial systems based on visible light wavelengths.
Remote acquisition of infrared images is also not
problematic (only the required optics are more expensive).
 Fingerprint systems use images of ridge patterns on
finger tips or hand palms [37]. Biometric systems use live
fingerprint scanners for fingerprint acquisition but people
leave fingerprints on anything they touch. Latent
fingerprints are in fact the fat (and some other substances)
left by the finger on the touched object. The best objects to
look for fingerprints are glass, doorknobs and glossy
paper [14]. There are several techniques to visualize and
digitalize the latent fingerprint. Both the forensic practice
and experiments [14] show that this is relatively simple. It
is said that on average people leave 25 perfect fingerprints
every day [47]. People normally have 10 fingers and 10
toes. Toes are not currently used for biometric purposes.
Little fingers often do not convey enough biometric
information for reliable identification/verification.
Typically left and right thumbs and index fingers are used
in fingerprint systems, the right index finger being the
most popular finger used. Most fingerprint-based
biometric systems will in fact use the same finger(s).
 Iris based systems use images of the iris. The iris is an
internal organ, but it is directly visible and can be
observed easily. For better imaging of brown irises the
near-infrared band is often used, but good quality images
can be obtained also in the visible part of the spectrum.
For reliable identification or verification the diameter of
the iris in a digital image of the eye should be at least 100
pixels [31]. A good quality camera and zoom lens can be
used to acquire an image of the iris with sufficient quality
even from mediate distances. Covert image capture is also
possible as the cooperation of the data subject is not
necessarily required. People have two eyes, typically both
eyes are enrolled and only the dominant eye is used for
identifications/verifications.
 Voice based systems use the information about the vocal
tract conveyed in the voice to identify/verify humans. The
voice based biometric systems can be divided into
text-dependent (always use the same phrase),
text-prompted (computer generated phrase is used) and
text-independent (data subject can pronounce any phrase)
systems. The data subject uses his vocal tract not only for
the biometric authentication but also during normal
conversation and these are both situations where an
attacker can record the data subjects's voice (overtly or
covertly). The situation of the attacker is different in
different system types. In the text-dependent systems all
the attacker needs is the fixed phrase which can be
user-specific or the same for all the data subjects. In
text-independent system any recording of the data subject
is sufficient. In text-prompted systems the attacker does
not know in advance what phrase will be required; still the
attacker can assemble the required phrase from several
Matyáš and Říha177
other recorded phrases or create a model of the speaker,
train the model with the available recordings of the data
subject and use the model to approximate the
pronunciation of the required phrase. This area is currently
open for research.
 Subskin biometric characteristics are not directly
visible and a special scanner (e.g., ultrasonic) is required
to get the biometric data. While completely covert
scanning is difficult, it may be possible to obtain the scan
by using some cover stories, as the scanning devices can
be easily masked as looking as something else. Usually
subskin fingernail characteristics are used for biometric
purposes.
 Authentication based on the mind [56] is measuring the
brain activity during the authentication process. To obtain
such biometric data we would have to use similar
measurement devices (difficult to do covertly) and make
the data subject to authenticate (to perform the same brain
activities, e.g. to think a password). The brain activity is
different for every combination of a person and a
password. It is difficult to estimate how hard it would be to
create a model of a human brain, train it with known
combinations of persons and passwords and then derive
the biometric data for another person or password.
 Retina based systems use the pattern of vessels behind
the retina in choroid to identify/verify humans. The retina
is an internal part of the eye and is not directly visible. To
obtain the image of blood vessels in choroid a special
instrument must be used. Ophthalmologists use devices
called retinoscopes, automated biometric systems use
specific optical devices to acquire the image of the blood
vessels. Covert retina scan is not possible, overt retina
scan without the cooperation of the data subject is also not
possible. People only have two eyes, so the biometric
systems based on retina usually share the same biometric
information.
As we can see there are substantial differences between
various systems. The order of biometric technologies in which
they are presented above reflects our opinion about the
difficulty of obtaining relevant biometric data (from easy to
get to more difficult to get). It is, however, only a rough
estimate and different people can have different opinions and
there are several factors which need to be taken into account
(like the type of the system in voice-based biometric
solutions). Some biometrics are not secret because their traces
are left all over the place. Others are potentially insecure as
they are not revocable and their use in multiple systems can
lead to problems with secrecy in case one copy of the
biometric sample or characteristic leaks.
IV. Template protection
Attacks similar to dictionary attacks in the password world
exist also in the biometric world – as for passwords, also not
all the biometric references are equally probable. Some
combinations of features are more likely than other.
Additionally, biometric data of some data subjects contain
significantly less features (than the average). Therefore some
biometric references can be ‘weaker’ than others, and so the
biometric dictionary attacks will focus on more probable and
weaker biometric references. Unlike in the password world,
there is not much that the data subjects can do to improve the
situation. The system designer can opt for a more restrictive
configuration, with both positive and negative effects of such
configuration.
Biometric dictionary attacks are only relevant to remote (and
API) authentication, as locally the liveness test should stop
non-genuine biometric samples. The problem of biometric
dictionary attacks can be mitigated similarly as in the case of
password dictionary attacks, for example by limiting the
number of unsuccessful trials or progressively extending the
timeout after unsuccessful attempts.
If a biometric system returns not only the YES/NO answer,
but also the resulting score, then this score return can be
misused. Using so called hill climbing [3, 53, 57] the score
can be used to progressively improve the biometric data being
evaluated until a successful comparison is achieved. It seems
that even quantization of the score does not help significantly
[2].
Other discussions relate to the format of the biometric
reference. Although there are several ISO standards (ISO/IEC
19794 series) defining open and interoperable biometric
reference formats, most of these formats are based on images.
The notable exception is the fingerprint minutiae format
(ISO/IEC 19794-2). Independent tests (the MINEX 2004 [26]
and MTIT [42]), however, concluded that the accuracy of
comparison algorithms is significantly better when the
algorithms use their own proprietary template formats.
Therefore systems that have strict demands on accuracy (and
prefer to use templates) still rely on proprietary template
formats. In some cases raw biometric data (e.g., the image) are
stored in addition to or instead of the template. The reasons
for doing so include the need of original data for legal proofs
or for interoperability between a number of system which do
not necessarily share the same biometric reference format
(e.g., electronic passports) and possibility to reuse the data in
an upgraded biometric system with an enhanced biometric
reference format. The storage of raw biometric data raises
privacy issues (the biometric data can contain some sensitive
personal information) and fears about easier misuse of the
data (in the case of compromise the full data is compromised
and better interoperability implies also easier crossmatching
between multiple databases).
Even in situations where only the template is stored (and not
the raw biometric data) such templates can be misused to
crossmatch several databases. Researchers are designing
schemes where two templates from two different databases
cannot be directly compared, such templates can only be used
for comparisons against raw biometric data (e.g., images).
The naïve solution is to encrypt the templates, but then the
templates need to be decrypted before the comparison. Some
more advanced solutions, like the use of shielding functions
[38, 54] or so called cancellable templates [49, 48, 40] can
help to limit the misuse of the templates (a closely related
subject of fuzzy vaults is discussed later in this paper). It is,
however, necessary to understand that having access to the
raw biometric data enables comparisons against any kind of
template, as this is the inherent property of biometric
templates. And all biometric systems need to process the raw
data (which can be potentially misused).
Security of Biometric Authentication Systems 178
V. The liveness problem
As we have discussed above, biometric data are not secret.
We cannot expect that the biometric characteristics are not
available to attackers. Therefore the knowledge and
presentation of the data subject's biometric data should not
directly lead to a successful authentication. This is why
remote biometric authentication does not work (is not secure)
in most cases (despite of some broad framework
specifications [33]). Locally we can try to verify so called
liveness (also called liveliness) to make sure we are
processing the fresh biometric data originating from the
person being authenticated.
Liveness tests are specific for a particular biometric modality
and can be roughly divided into two categories. Static tests
measure some physiological characteristics (like the finger
temperature or conductivity) that should discriminate between
the living human and an artificial fake. Dynamic tests verify
the reaction of the person to an impulse. The impulse can be
the increase in light intensity to see the pupil contraction or
asking the subject to pronounce a particular phrase.
There are many various kinds of liveness tests used today. For
example to verify the liveness of the fingerprints it is possible
to measure the temperature, reaction to hot and cold stimulus
or pressure stimulus, conductivity and other electrical
properties [36], the perspiration [19], optical properties of the
skin, contact scattering [9], pulse or blood oxygenation. We
can use hippidus effect in the case of the iris liveness test,
reaction to the volume or position of the illumination source
or look at the Fourier plane to deal with colour contact lenses
[18]. Facial recognition systems either use several cameras to
obtain 3D properties of the head (to avoid simple attacks with
a photo) or ask the subject for a particular reaction (to blink, to
move left or right, to open or close the mouth). Text prompted
voice systems ask the subject to pronounce a random phrase.
A combination of facial and voice recognition can also verify
the lip movement.
Unfortunately, most liveness tests are weak and can be easily
fooled by using materials having the same properties as the
human body or can be simulated in other ways. For the
discussion how easy it is to fool common fingerprint scanners
with silicon or gelatine copies (so called ‘gummy fingers’) see
[41, 14]. It is also necessary to mention that liveness tests
cannot detect changes of biometric characteristics (e.g., a
plastic surgery). It is by no means trivial to come up with a
liveness test that would be difficult to fool and would not
cause a high FRR of the entire system. From the medical point
of view, liveness would primarily be tested by means of the
EEG. But that is not exactly what we need in liveness testing
for biometric authentication systems. It is not sufficient to
measure the brain activity of the subject to verify that she is
indeed alive when she could use a plastic layer on her finger to
fool the scanner. The liveness test is therefore not only about
measuring the liveness of the subject, but more-or-less about
resistance to attacks with non-genuine biometric samples
(sometimes the term liveness+ is used). According to our
experience and private discussions with experts in the field
you can always bypass the liveness test if you know what the
liveness test is looking for. This is also the reason why details
of many liveness tests are kept secret and evaluation of their
security is not possible or at least not straightforward [39].
Many scanners were easily fooled with simple biometric
copies in the past. There are, however, a few biometric
sensors which are believed to be moderately spoof-resistant.
One of such sensors is the Lumidigm fingerprint sensor that
combines optical fingerprint scanner with multispectral
imaging [50]. No independent evaluation of the technology is
publicly available, but the US Transportation Security
Administration (TSA) tested the sensor and approved it in
2007. Currently (2010) the biometric Qualified Product List
(QPL) [29] lists 2 Lumidigm, 1 Bioscript, 1 Cogent and 1
Integrated Biometrics scanners.
VI. Secure biometric authentication
We have discussed the secrecy of biometric data and the need
for liveness testing. For secure authentication the biometric
system must be convinced that the presented biometric
measurements are coming from a trusted and unmodified
input device and are fresh. The biometric system should
verify the liveness, otherwise the system could be cheated
with copies of biometric characteristics. Sometimes the
liveness test can be replaced with a human guard, it is
however questionable whether a human guard can protect
against more advanced biometric fakes (like a thin silicon
layer at the fingerprint).
If the authentication is done on-device, the device itself
should be trustworthy. If the authentication is done
off-device, then the operating environment of the software
and the communication link between the software and the
device have to be secured. For example, in a client-server
application, if the client workstation is not trusted, then there
is no point authenticating the person using that workstation. If
one chooses to run the authentication software at the server
side, then the communication link between the server and the
device itself (not just the client workstation) has to be secured.
Otherwise, a malicious party or even the workstation itself
may intercept the communication and replay recorded
biometric data. One way to defeat replay attacks is to put a
separate secret key in the device and to use
challenge/response protocol with this key. Obviously, the
device has to be trustworthy. One possible solution would be
to use a tunnelling protocol with mandatory authentication of
both parties. To protect the keys and to avoid modification of
the liveness test the device must be tamper-resistant or
physically secured.
As we have already mentioned, remote biometric
authentication is mostly not secure. There is no sense to send a
fingerprint to log-in to a web server if the fingerprint scanner
is not trusted by the web server. And to be trusted can imply to
have a reliable liveness test, to have a secret key to support
data authenticity and to be tamper resistant (to protect the keys
and to protect against modification of the liveness test or
direct injection of attacker's data – the device is fully in the
hands of users and potential attackers). It is quite difficult to
make a small smartcard tamper resistant. To design a tamper
resistant fingerprint scanner is a real challenge.
The raw biometric sample or templates need to be supplied for
comparisons. Hashing the biometric data by a cryptographic
hash function and sending only the hash [13] for comparison
does not work (unlike for passwords). Biometric
measurements never yield the same values, therefore cannot
be directly compared in hashed or encrypted domain. There
are some efforts of how to avoid the transmission of the full
sensitive biometric data [11] and we discuss relevant issues
below.
Matyáš and Říha179
VII. Biometric encryption
The idea of cryptographic keys derived exclusively from
biometric data bears some very attractive advantages, e.g., the
keys being used only with the rightful owner present (and
being re-generated ‘on the fly’, then could be destroyed after
use).
However, such derived keys also have some unpleasant
properties that make them useless in many traditional
cryptographic applications [5]. Such keys have limited
entropy [46], are created through a deterministic process from
non-secret biometric data, and also cannot be changed (or
only several times). Invariant features can obviously be
extracted from a biometric sample and encoded so that they
can be used as an encryption key. The basic problem with this
concept is that such a key cannot be treated in the same way
cryptographic keys are usually treated. Here we shall again
stress the fact that biometrics are not secret. Other factors may
include the volume of data that is truly invariant over time and
the problems with changing one’s key as soon as the data
subject actually finds out that her fingerprint has been
disclosed. Moreover, a damaged fingerprint would disable
any operations with the derived key.
Biometric encryption (together with other methods of
template protection like cancellable biometrics and shielding
functions) belong to the area called renewable biometrics and
has been already studied for some time. The emerging ISO
standard 24745 Information technology – Biometric
information protection [30] (currently, November 2010, in its
FCD – Final Committee Draft) is unifying the terminology,
explaining the basic principles and refers to example
technologies [12].
A. Entropy of biometric characteristics
While unprocessed biometric samples have the size of
kilobytes or megabytes (e.g., the scan of a hand palm in a high
resolution), the entropy of the repeatable invariant biometric
features is much smaller. The estimation of the entropy of
biometric characteristics has recently become an area of
active research [1, 4, 59]. The entropy estimates cannot
evaluate the entropy of general biometric samples, but are
specific for biometric modality (fingerprints) and the
particular features used for identification (fingerprint
minutiae). It should be noted that it is hard to pin down the
entropy in a precise way. A. Adler et al. [1] estimate the
entropy of facial biometric information to 40-50 bits. M.
Young [59] estimates the entropy of fingerprint minutiae to 82
bits. The iris template (in Daugman based algorithms) is 2048
bits long. When combined with error correction codes the
iriscode is able to produce repeatable strings of 140 bits with a
low false rejection rate (experiments produced bitstrings of
42-224 bits with various error rates, for details see [27]).
Other estimates quote 250 bits of entropy per iriscode [34] or
260 bits per iris image [60].
Interestingly, the noisy part of raw biometric data that is
ignored by biometric algorithms (when searching for the
invariant features) can be used to generate random numbers
[24].
B. Secrecy and changeability of biometric data
Important properties of cryptographic (symmetric and private
asymmetric) keys include secrecy of the keys and
changeability after a key compromise. As we have discussed
above, biometric data are neither secret nor changeable. This
means that ‘keys’ derived by publicly available algorithms
from non-secret data cannot be considered secret and do not
fulfil this critical requirement of cryptographic keys.
To improve secrecy of the resulting data we can either make
the algorithm secret (but security-by-obscurity is not a good
design principle) or make the result a function of not only the
biometric data, but other secret data as well (password or
secret key). Then the biometric data are only one of several
authentication factors.
C. Repeatability of biometric data
One of the most serious problems any implementation of
biometric key derivation has to face is the variability of the
biometric data. The raw biometric samples are never the same,
they only are similar. In biometric authentication, we need a
metric of such a similarity. For encryption, we have to derive
exactly the same bitstring as we know that a single changed
bit will radically change the results of cryptographic
operations (as a consequence of the required avalanche
effect). Deriving exact repeatable bitstrings from noisy
biometric data is not a simple task.
One of the possible options is to use discretization with a
sufficient margin for errors (e.g., only 1 bit per feature which
is normally measured in several bits). Another approach is to
use error correction codes. The problem with biometric data is
the correlation of errors (a whole part of the fingerprint
missing, not just random features). To solve such difficulties a
combination of error correction codes can be used [27]. The
theoretical work in this field includes the notion of fuzzy
extractor [21] (error correction followed by hashing) and
fuzzy commitment scheme (a standard random key is
generated, then redundancy for error correction is added and
result is XORed with the biometric data) [35].
Relevant to this field are also advances in threshold
cryptography (secret sharing, etc.) [20]. It is also worth
mentioning the work carried out by Wheeler [58], which
focuses on the agreement on a key using faulty (analogue
biometric) data from separate measurements, where
error-correction is involved and at the end of such protocol
execution both the sender and the receiver have the same
bitstring. Another work [16] focuses on MACs of noisy data
like biometrics.
D. ‘Biometric keys’
There are several types of keys, which are somehow protected
by biometrics. According to how exactly biometric data is
applied to get the so-called biometric key, we can divide the
systems into three categories: systems with key unblocking,
key derivation and key locking.
1) Key unblocking
The first option where biometrics can meet cryptography is
securing access to a secret key. Access to secret keys is
commonly protected by passwords or PINs. Securing the
secret key also with a biometric system might improve its
protection. In such a case, the cryptographic key is a standard
(random) key and it is released only after successful biometric
authentication.
If biometric measurements are not acquired by the signing
device itself, the signing device must verify that the data were
really captured by an authentic biometric sensor and are fresh,
otherwise the use of biometrics does not provide much
security.
Security of Biometric Authentication Systems 180
The protection of the cryptographic key by biometrics
requires the use of a trusted component (a trusted workstation
or a secure hardware). An attacker cannot have a direct access
to the biometric reference and the key, otherwise the
biometric authentication can be bypassed completely and the
cryptographic key can be recovered directly without any
authentication.
Key unblocking is not a proper biometric key scenario rather
than a kind of biometric authentication.
2) Derivation of encryption keys directly from biometric
data
Another option is to derive the biometric key solely from the
biometric data. Here the principal engineering problem is the
variability of the biometric samples. Such noisy data have to
be transformed into a repeatable bitstring. Discretization,
error correction codes and other techniques can be used to
help to implement such a scheme.
There have been several attempts to implement such schemes.
One of the first systems was based on keystroke dynamics
[44]. Biometric information (here in the terms of the
key-down time and the inter-key delay) was added to the
typed password after discretization. Such ‘hardened
password’ was the resulting biometric key. Another approach
used voice biometric data to generate the key [43]. The system
was also based on discretization. Handwritten signatures can
also produce a biometric key [28]. Many encoded dynamic
properties like the velocity, pressure and direction are
concatenated to form the resulting key.
All these implementations suffer from high false rejection
rates (the legitimate data subject is unable to reproduce the
original key) in the range of 20-50%, which makes them
practically unusable. Even with so high FRR the length of the
resulting biometric key is quite short (being in the range of
12-46 bits).
Keys derived directly from biometric data are of limited use
no matter what type of key is generated (symmetric,
asymmetric) or the kind of application the key is intended for
(digital signatures, encryption and decryption of documents).
This relates to the fact that there is no secret information
involved in the entire process. Such biometric keys can be
considered as a form of a biometric template.
3) Key locking
The third approach of cryptography meeting biometrics
combines the first two methods. Unlike during the key
unlocking the data subject's biometric data is applied to a
random cryptographic key. The resulting locked key should
not leak information neither about the original biometric data
nor about the cryptographic key. Only the correct biometric
data (biometric sample similar to the original one – practically
a sample originating from the same person) should be able to
unlock the key and produce the same cryptographic key to
which the biometric data was applied at the beginning [6].
Several implementations have tried to follow such a scheme.
A fingerprint based implementation [52] called Bioscript used
phase information of the Fourier transformation of fingerprint
images and majority coding for locking the randomly
generated key (so-called biometric locking). Another
fingerprint implementation [15] used minutiae points (plus
other chaff points) to create a locking set, from which a secret
key could be recomputed. The technique is called the fuzzy
vault. Another implementation based on biometric locking
appeared in the area of facial biometric systems [25]. Here the
facial eigen projections are used as the primary biometric data
and majority coding and polynomial thresholding are used in
the process of the key generation. These implementations also
suffer from high FRRs and short resulting bitstrings.
Another implementation [27] is based on iris biometric data.
First a random 140-bit cryptographic key is taken and
encoded using Reed-Solomon and Hadamard codes to form a
2048-bit string. This bitstring is then locked by XORing it
with the data subject's iriscode. The result is called
pseudoiriscode. The process of decoding XORs the
pseudoiriscode with a fresh iriscode and applies the
Hadamard and Reed-Solomon decoding to obtain the original
140-bit key. The authors claim the FRR to be 0.47% and FAR
to be 0% (for the 140-bit keys).
In fact any algorithm reliably deriving a biometric key from
the biometric data can be enhanced by XORing the key with a
random key to design a key locking scheme. Also locked
biometric keys have to cope with the low secrecy of biometric
data as any biometric sample of the correct data subject can be
used to unlock the key. Therefore the locking cannot be
considered as a sufficient protection and the locked key must
be kept secret.
The main advantage is changeability of the cryptographic key
which is being locked. The recovery after a key compromise
is easy. On the other hand if the biometrics are compromised,
there is no easy recovery and the same biometric data can
unlock next locked keys as well (unless other biometrics –
like another finger for fingerprint systems – is used).
VIII. Conclusions
The primary advantage of biometric authentication methods
over other methods of user authentication is that they really do
what they should, i.e., they do authenticate the user. They do
not rely on objects the user carries or something the user has
remembered. Biometric authentication methods use the real
human physiological or behavioural characteristics to
authenticate users. These characteristics should not be
duplicable, but it is unfortunately often possible to create a
copy that is accepted by the biometric system as a true sample.
Well-known investigations [41, 55] confirmed our earlier
findings that attacks may be much easier than generally
accepted.
The biometric authentication also has some other advantages.
Most biometric techniques are based on something that cannot
be lost or forgotten. This is an advantage for users as well as
for system administrators because the management of lost,
reissued or temporarily issued tokens/cards/passwords can be
avoided.
So why do not we use biometrics everywhere instead of
passwords or tokens? Nothing is perfect and biometric
authentication methods also have their own shortcomings.
First of all the performance of biometric systems is not ideal
(yet?). Getting the result is quite clear and quick when
comparing two passwords. Comparing two sets of biometric
characteristics is not so straightforward. Biometric systems
still need to be improved in the terms of accuracy (and
sometimes also speed).
The fail to enrol rate (FTE) brings another important problem.
Not all users can use any given biometric system. People
without hands cannot use fingerprint or hand-based systems.
Visually impaired people have difficulties using iris or retina
Matyáš and Říha181
based techniques. Even enrolled users can have difficulties
using a biometric system.
Biometric data are not secret and the security of a biometric
system cannot be based solely on the secrecy of the biometric
characteristics. The server cannot authenticate the person just
after receiving her correct biometric characteristics. User
authentication can be successful only when the person’s
characteristics are fresh and have been collected from the
person being authenticated. This implies that the biometric
input device must be trusted. Its authenticity should be
verified (unless the device and the link are physically secure)
and the liveness should be checked.
We believe that the biometric authentication is a good
additional authentication method. Even cheap and simple
biometric solutions can often increase the overall system
security if used on top of existing traditional authentication
methods. Replacing a current traditional authentication
method with a biometric one, on the other hand, may be risky
and requires deeper analyses. Biometric authentication
systems often replace traditional authentication systems not
because of their higher security but because of higher comfort
and ease of use.
Biometric key generation is far from mature. The high false
rejection rate and short key length are the common
shortcomings of most current systems, yet the first practically
usable implementations are appearing. However, the use of
biometric keys has to cope with the fact that biometric data are
not (fully) secret. This makes the use of directly derived
biometric keys very problematic, but also biometrically
locked cryptographic keys have to take into account that
compromise of the biometric data implies easier access to the
locked keys – and prudent implementers will protect the
secrecy of the locked keys by additional means.
References
[1] Adler, A., Youmaran, R., Loyka, S. (2009) ‘Towards a
Measure of Biometric Information’ in Pattern Analysis
& Applications, Volume 12, Number 3, Springer
London, September 2009.
[2] Adler, A. (2004) ‘Reconstruction of source images from
quantized biometric match score data’ at Biometrics
Consortium Conference 2004, Washington, DC, USA,
September 2004.
[3] Adler, A. (2004) ‘Vulnerabilities in biometric encryption
systems’, at NATO RTA Workshop: Enhancing
Information Systems Security – Biometrics,
IST-044-RWS-007.
[4] Adler, A., Youmaran, R., Loyka, S. (2005) ‘Information
content of biometric features’, at Biometrics Consortium
Conference 2005, Washington, DC, USA, September
2005.
[5] Ballard, L., Kamara, S., and Reiter, M. K. (2008) ‘The
practical subtleties of biometric key generation’, in
Proceedings of the 17th Conference on Security
Symposium, San Jose, CA, USA, USENIX Association.
[6] Ballard, L., Kamara, S., Monrose, F., Reiter, M. K.
(2008). ‘Towards practical biometric key generation
with randomized biometric templates’, in Proceedings
of the 15th ACM Conference on Computer and
Communications Security, Alexandria, Virginia, USA.
ACM, New York, NY, USA.
[7] Bausinger, O., Seidel, U. (2007) ‘Next Generation
German e-Passport Fingerprint Enrolment - Quality vs.
Time’ at NIST Biometric Quality Workshop 2007, NIST,
2007.
[8] UK Passport Service (2005) ‘Biometrics, enrolment
trial’, Management Summary.
http://www.ips.gov.uk/cps/files/ips/live/assets/documen
ts/UKPSBiometrics-Enrolment-Trial-Report-Managem
ent-Summary.pdf
[9] Bicz, W. (2003) ‘The Impossibility of Faking Optel’s
Ultrasonic Fingerprint Scanners’, February 2003,
http://www.optel.pl.
[10]Biometrics Working Group (2000) ‘Best Practices in
Testing and Reporting Performance of Biometric
Devices’.
[11]Boyen, X et al (2005) ‘Secure Remote Authentication
Using Biometric Data’ in Advances in cryptology –
EUROCRYPT 2005, pp. 147–163.
[12]Breebaart, J., Yang B., Buhan-Dulman I., Busch Ch.
(2009) ‘Biometric template protection’ in Datenschutz
und Datensicherheit - DuD, Volume 33, Number 5,
Vieweg Verlag,.
[13]Calabrese, C. (1999) ‘The trouble with biometrics’ in
;login:, Vol 24, No 4, ISSN 1044-6397.
[14]Chaos Computer Club (2004) ‘How to fake
fingerprints?’.
http://dasalte.ccc.de/biometrie/fingerabdruck_kopieren?
language=en.
[15]Clancy, T.C., Kiyavash, N., Lin, D.J. (2003) ‘Secure
Smart Card-Based Fingerprint Authentication’ in Proc.
2003 ACM SIGMM Workshop Biometrics Methods and
Application, 2003.
[16]Crescenzo, G. et al. (2005) ‘Approximate Message
Authentication and Biometric Entity Authentication’ in
Financial Cryptography and Data Security, LNCS
3570, 2005.
[17]Daugman, J. (1998) ‘Phenotypic versus genotypic
approaches to face recognition’, in Face Recognition:
From Theory to Applications, Heidelberg,
Springer-Verlag, ISBN 3-540-64410-5.
[18]Daugman, J., ‘Anti-spoofing Liveness Detection’.
Exchanged by personal communication in April 2001.
[19]Derakhshani, R., Schuckers, S., Hornak, L., O'Gorman,
L. (2001) ‘Determination of vitality from a non-invasive
biomedical measurement for use in fingerprint scanners’
in Pattern Recognition Journal, Vol. 36, pp. 383-396,
December 2001.
[20]Desmedt, Y. (1997) ‘Some Recent Research Aspects of
Threshold Cryptography’ in Information Security,
Proceedings, LNCS 1396, pp. 158-173,
Springer-Verlag.
[21]Dodis, Y., Reyzin, L., Smith, A. (2004) ‘Fuzzy
extractors: How to Generate Strong Keys from
Biometric and Other Noisy Data’ in Proc. Eurocrypt
2004, pp. 523-540, Springer-Verlag, 2004.
[22]FRONTEX (2007) BIOPASS – Study on Automated
Biometric Border Crossing Systems for Registered
Passenger at Four European Airports. European Agency
for the Management of Operational Cooperation at the
External Borders of the Member States of the European
Union. 2007.
[23]FRONTEX (2010) BIOPASS II – Automated biometric
border crossing systems based on electronic passports
Security of Biometric Authentication Systems 182
and facial recognition: RAPID and SmartGate.
European Agency for the Management of Operational
Cooperation at the External Borders of the Member
States of the European Union. 2010.
[24]Gerguri, S., Matyas, V., Riha, Z., Smolik, L. (2010)
‘Random Number Generation Based on Fingerprints’ in
Information Security Theory and Practices: Security
and Privacy of Pervasive Systems and Smart Devices.
Berlin, Springer, 2010.
[25]Goh, A., Ngo, D.C. L. (2003) ‘Computation of
Cryptographic Keys from Face Biometrics’, in Proc. 7th
IFIP TC6/TC11 Conf. Commun. Multimedia Security,
pp. 1-13, Springer-Verlag, LNCS 2828, 2003.
[26]Grother P. et al. (2006) ‘MINEX, Performance and
Interoperability of the INCITS 378 Fingerprint
Template’, NISTIR 7296.
[27]Hao, F., Anderson, R., Daugman, J. (2006) ‘Combining
Crypto with Biometrics Effectively’ in IEEE
Transactions on Computers, Vol. 55, No. 9, 2006.
[28]Hao, F., Chan, C.W. (2002) ‘Private Key Generation
from On-Line Handwritten Signatures’ in Information
Management & Computer Security, Vol. 10, No. 2, pp.
159-164, 2002.
[29]Hendricks R. (2010) ‘Airport Biometrics Access Control
Qualified Products Lists (QPL)’ at Qualifying Identity
and Privilege Credential Products for the
Transportation Worker Identification Credential
(TWIC) and other DHS/TSA programs Qualified
Product Lists (QPL), NIST, Gaithersburg, MD, USA,
2010.
[30]ISO/IEC JTC 1/SC 27 (2010) ‘ISO/IEC FCD 24745
Information technology – Biometric information
protection’.
[31]ISO/IEC JTC 1/SC 37 (2004) ‘ISO 19794-6 Biometric
Data Interchange Formats – Part 6: Iris Image Data’.
[32]ISO/IEC JTC1/SC 37 (2006) ‘Harmonized Biometric
Vocabulary’.
[33]ITU-T (2004) ‘X.1081 - The Telebiometric Multimodal
Model - A Framework for the Specification of Security
Aspects of Telebiometrics’.
[34]Juels, A. (2004) ‘Fuzzy Commitment’ at DIMACS
Workshop on Cryptography: Theory Meets Practice,
October 2004.
[35]Juels, A., Wattenberg, M. (1999) ‘A Fuzzy Commitment
Scheme’ in Proc. Sixth ACM Conf. Computer and
Comm. Security, 1999.
[36]Kallo, P., Kiss, I., Podmaniczky, A. (2001) ‘Detector for
recognizing the living character of a finger in a
fingerprint recognizing apparatus’, US patent 6175641,
January 2001.
[37]Krishneswari, K., Arumugam, A. (2010) ‘A Review on
Palm Print Verification System’, International Journal of
Computer Information Systems and Industrial
Management Applications (IJCISIM), ISSN: 2150-7988
Vol.2 (2010), pp.113-120.
[38]Linnartz, J.-P., Tuyls P. (2003) ‘New shielding functions
to enhance privacy and prevent misuse of biometric
templates’ in Proc. 4th Int. Conf. AVBPA, pp. 393–402,
2003.
[39]Liveness Detection in Biometric Systems,
http://www.biometricsinfo.org/whitepaper1.htm.
[40]Maiorana, E. (2010) ‘Biometric cryptosystem using
function based on-line signature recognition’ in Expert
Systems with Applications: An International Journal,
April 2010.
[41]Matsumoto, T., Matsumoto, H., Yamada, K., Hoshino, S.
(2002) ‘Impact of Artificial ‘Gummy’ Fingers on
Fingerprint Systems’, SPIE Vol. #4677, Optical
Security and Counterfeit Deterrence Techniques IV,
January 2002, http://cryptome.org/gummy.htm.
[42]MTIT, Minutiae Template Interoperability Testing,
http://www.mtitproject.com.
[43]Monrose, F., Reiter, M.K., Li, Q., Lopresti, D.P., Shih, C.
(2002) ‘Toward Speech-Generated Cryptographic Keys
on Resource Constrained Devices’, in USENIX Security
Workshop 2002.
[44]Monrose, F., Reiter, M.K., Wetzel R. (1999) ‘Password
Hardening Based on Keystroke Dynamics’ in Proc.
Sixth ACM Conf. Computer and Comm. Security.
[45]Phillips P. J. et al (2007) ‘FRVT 2006 and ICE 2006
Large-Scale Results’, http://www.frvt.org/.
[46]Plaga R. (2009) ‘Biometric keys: suitable use cases and
achievable information content’ in International Journal
of Information Security, Volume 8, Number 6, Springer
Berlin / Heidelberg, 2009.
[47]van der Putte, T., Keuning, J. (2000) ‘Biometrical
Fingerprint Recognition Don't Get Your Fingers
Burned’ in Fourth Working Conference On Smart Card
Research and Advanced Applications, pp. 289-303,
Kluwer.
[48]Ratha N.K., Chikkerur S., Connell J. H., Bolle, M. B.
(2007) ‘Privacy Enhancements for Inexact Biometric
Templates’ in Security with Noisy Data, Springer
London.
[49]Ratha N.K., Chikkerur S., Connell J. H., Bolle, M. B.
(2007) ‘Generating Cancelable Fingerprint Templates’
in IEEE Transactions on Pattern Analysis and Machine
Intelligence, Volume 29, Issue 4, IEEE Computer
Society, Washington, DC, USA, 2007.
[50]Rowe, R. (2005) ‘A Multispectral Sensor for Fingerprint
Spoof Detection’ in SENSORS, 22(1), p. 1-4., January
2005.
[51]Seidel U. (2009) ‘Fingerprint capture and the German
Experience’ in MRTD Report, Number 2, Volume 4,
International Civil Aviation Organization.
[52]Soutar, C., Roberge, D., Stoianov, A., Gilroy, R., Vijaya
Kumar, B.V.K. (1999) ‘Biometric Encryption’ in ICSA
Guide to Cryptography, McGraw-Hill, 1999.
[53]Soutar, C. ‘Biometric System Security’, White Paper,
bioscrypt, http://www.bioscrypt.com/.
[54]Tuyls, P. et al. (2005) ‘Practical Biometric
Authentication with Template Protection’ in AVBPA
2005, LNCS 3546, pp. 436-446, 2005.
[55]Thalheim, L., Krissler, J., Ziegler, P.-M. (2002) ‘Body
Check’ in c’t magazine 11/2002.
[56]Thorpe, J., van Oorschot, P.C., Somayaji, A. (2005)
‘Pass-thoughts: Authenticating With Our Minds’, in
Proceedings of the ACSA 2005 New Security Paradigms
Workshop, Lake Arrowhead, California, USA,
September 2005.
[57]Uludag, U, Jain, A.K. (2004) ‘Attacks on biometric
systems: a case study in fingerprints’, in Proc. SPIE-EI
2004, pp. 622-633, San Jose, CA, January 2004.
[58]Wheeler, D. (2001) ‘Protocols Using Keys from Faulty
Data’, in Security Protocols Workshop 2001,
Springer-Verlag LNCS 2467.
Matyáš and Říha183
z
[59]Young, M. (2006) ‘Equating Biometric Entropy’ in
Biometric Consortium Conference, Baltimore, MD,
2006.
[60]Ziauddin S., Dailey M.N. (2010) ‘Robust iris verification
for key management’ in Pattern Recognition Letters,
Elsevier, 2010.
Author Biographies
Václav (Vashek) Matyáš is a Professor at the
Masaryk University Brno, CZ, chairing its
Department of Computer Systems and
Communications. His research interests relate to
applied cryptography and security, where he
published nearly hundred peer-reviewed papers and
articles, and co-authored six books. He worked with
Microsoft Research Cambridge, University College
Dublin, Ubilab at UBS AG, and was a Royal Society
Postdoctoral Fellow with the Cambridge University Computer Lab. Vashek is
one of the Editors-in-Chief of the Identity in the Information Society journal,
he edited the Computer and Communications Security Reviews, and worked
on the development of Common Criteria and with ISO/IEC JTC1 SC27. He
received his PhD degree from Masaryk University, Brno and can be contacted
at matyas@fi.muni.c
Zdenek Riha is an Assistant Professor with the
Masaryk University in Brno, Czech Republic,
working in the area of computer security. He received
a PhD degree in Computer Science from the Faculty
of Informatics, Masaryk University in Brno. Between
2005 and 2008 he was seconded as National Detached
Expert to the Joint Research Centre of the European
Commission. His interests include authentication,
identity documents, PKI and security of operating
systems.
Security of Biometric Authentication Systems 184