PATTERNS 2012 : The Fourth International Conferences on Pervasive Patterns and Applications

Iris Recognition: Existing Methods and Open Issues

Sajida Kalsoom Sheikh Ziauddin
Department of Computer Science Department of Computer Science
COMSATS Institute of Information Technology COMSATS Institute of Information Technology
Islamabad, Pakistan Islamabad, Pakistan
Email: [email protected] Email: [email protected]

AbstractBiometric authentication uses unique physical or segmentation, iris normalization and iris feature encoding.
behavioural patterns in humans to identify individuals. Though The iris segmentation procedure segments the annular iris
biometric is generally considered most reliable, stable and unique region from the entire eye image. First, it finds the inner
among all entity authentication means, it is not as stable and
unique as is usually conceived. In this paper, we highlight the and outer boundaries (the iris-pupil and iris-sclera boundaries)
issues with current state-of-the-art iris-based biometric authen- of the iris, then it marks the region of the annular iris ring
tication systems. This survey covers the review of existing iris that is not visible due to eyelids and eyelashes. The iris
recognition methods with a focus on enumerating the open issues normalization procedure transforms the segmented iris region
that must be addressed in order to be more confident in the into a fixed size to cater for variations in iris sizes among
performance, security and privacy aspects of iris-based biometric
systems. different eye images. The feature encoding procedure extracts
the most distinguishing features from normalized iris images
Keywords-Pattern recognition; biometric authentication; iris and typically encodes the result as a binary string.
recognition; template security
Recognition involves either verification or identification.
I. I NTRODUCTION Verification is one to one comparison where claim of an
identity is verified, e.g., an employee of an office. On the
With the increase in use of biometrics for human identi- contrary, identification is one to many comparison where an
fication, control shifts to identifying the factors that affect identity is watched against an entire database, e.g., a criminal
the performance of biometric authentication systems. Bio- surveillance system. In the verification step, the recognition
metric authentication systems use behavioural or physical time captured image is compared with the image taken at the
characteristics to authenticate a user. These systems have enrolment time. The comparison is mostly done by calculating
become more reliable sources of authentication as compared the Hamming distance where a value of 0 represents a perfect
to the traditional means like passwords or hardware tokens match and a value of 1 represents perfect non-match.
such as smart cards. Reliability of biometric authentication This paper is not primarily a survey on iris recognition
systems lies in the fact that, unlike passwords and smart cards, techniques, but is to identify performance and security related
biometrics cannot easily be forged, shared, compromised or issues with existing techniques. So, we will briefly describe
forgotten. Biometric is considered to be highly unique among just a couple of representative systems, followed by a table,
all human population. Genetically, same identities including reproduced from [5], providing a comparative analysis of a
twins and irises of left and right eye of the same person rep- number of state-of-the-art iris recognition systems.
resent different iris patterns [8]. Another important property of The most famous iris biometric system is due to Daugman
biometric is its stability [23][24][25][34]. In this paper, we will [7][8]. In Daugmans system, iris segmentation is performed
critically analyse these claims by showing counter-examples by using the following optimization:
from other researchers work. These will be discussed in the
problems and open issues section in detail.
I
I(x, y)
max G (r) ds ,

The rest of the paper is organized as follows. Section II (r,x0 ,y0 ) r C(s;r,x0 ,y0 ) 2r
provides an overview of existing iris recognition methods.
Section III presents issues, problems and challenges associated where r and (x0 , y0 ) are candidates for the radius and center of
with template security and recognition performance. The paper the iris, G (r) is the one-dimensional Gaussian with standard
is concluded in Section IV. deviation , is the convolution operator, C(s; r, x0 , y0 ) is the
circular closed curve with center (x0 , y0 ) and radius r, param-
II. I RIS R ECOGNITION eterized by s, and I(, ) is the input eye image. Noise due to
Iris recognition is considered as one of the most reli- eyelids occlusion is avoided by restricting ds to the nearly
able biometric authentication technique [9][19][35][37]. An iris vertical regions. The above optimization is performed twice
recognition system captures human eye image using a near to find both iris and pupil circles. For template generation,
infrared iris sensor which passes through three steps to be Daugman uses phase information of the image. After detecting
transformed into an iris template. These three steps are iris the iris boundaries and removing the irrelevant region, 2D

Copyright (c) IARIA, 2012. ISBN: 978-1-61208-221-9 23

PATTERNS 2012 : The Fourth International Conferences on Pervasive Patterns and Applications

TABLE I
Gabor wavelets is applied on normalized iris image to the C OMPARISON OF THE MOST CITED IRIS RECOGNITION TECHNIQUES [5]
iris template. For comparison of iris templates, Hamming
distance metric is used. Most of the subsequent work on iris First Author, Year Database Size Results
recognition, follows Daugmans approach of using Hamming Alim, 2004 Not given 96.17%
distance for template matching
Jang, 2004 1694 images including 99.1%
After Daugmans iris recognition system, one of the most
160 w/glasses and
important and popular systems is due to Wildes [38][39]. For
iris segmentation, Wildes first detects edges in the eye image 11 w/contact lenses
and then applies a circular Hough transform to find circular Krichen, 2004 700 visible-light FAR/FRR: 0%/0.57%
pupil and iris boundaries. Much of the subsequent work on images
iris segmentation follows Wildes approach where a common Liu, 2005 4249 images 97.08%
variation is the usage of a coarse-to-fine strategy. For template Ma, 2002 1088 images 99.85%, FAR/FRR:
generation, Wildes uses Laplacian of Gaussian filter applied at
0.1%/0.83%
multiple scales to extract unique information from iris texture.
Ma, 2003 2255 images 99.43%, FAR/FRR:
His system uses normalized correlation between the templates
for template matching at verification time. 0.1%/0.97%
As mentioned above, most of the subsequent work in iris Ma, 2004 2255 images 99.60%, EER: 0.29%
recognition follows the above-mentioned seminal approaches. Ma, 2004 2255 images 100%, EER: 0.07%
Most work on iris segmentation is a variation and enhancement Monro, 2007 2156 CASIA images 100%
of Wildes approach, while most feature extraction schemes and 2955 U. of Bath
are variations on Daugmans wavelet-based approach. A very
images
nice detailed survey of iris recognition techniques is due to [5].
Proenca, 2007 800 ICE images EER: 1.03%
Table I (reproduced from [5]) provides a quick comparison of
recognition results for some of the important iris recognition Rossant, 2005 149 images 100%
techniques. The interested reader is referred to [5] for a Rydgren, 2004 82 images 100%
detailed study of existing iris recognition techniques. Sanchez-Reillo, 2001 200+ images 98.3%, EER: 3.6%
Son, 2004 1200 images, (600 used 99.4%
III. P ROBLEMS AND O PEN I SSUES
for training)
We categorize issues of iris recognition systems in two
Sun, 2004 2255 images 100%
broad classes namely those related to iris template security and
those associated with iris recognition performance. Details are Takano, 2004 Images from 10 people FAR/FRR: 0%/26%
as follows. Thornton, 2006 CMU database, 2000+ EER: 0.23%
images
A. Iris Template Security
Thornton, 2007 CMU database, 2000+ EER: 0.39%
As biometric is an integral part of human body, loss of ones images
biometric corresponds to loss of his/her identity. Therefore,
Tisse, 2002 300+ images FAR/FRR: 0%/11%
security of biometric templates is one of the most important
Yu, 2006 1016 images 99.74%
concerns in any biometric authentication system. In literature,
we found four types of biometric systems which are described
below along with related issues and challenges.
1) Traditional Biometric Systems: These are the conven- in a secure manner. Modern cryptographic keys are uniformly
tional systems [7][25][38] which store users templates in clear random and large in size, therefore it is not feasible for
form to verify the identity. A template is generated at enrol- users to memorize them. In biometric key release systems,
ment time, stored in the database without encryption/hashing cryptographic keys are stored at some location and are released
and compared with the corresponding verification template using biometric information of the user. When user inputs
at the verification time. As the template is used and stored his/her biometric, cryptographic key is released for use in any
in plaintext, a compromise of database has severe security security protocol. This way, the key would be released only
and privacy implications. There are scenarios where users to the authorized users.
use the same biometrics for multiple applications or different Though these systems use biometric information effectively
organizations share data among themselves for their users. In for cryptographic key storage and release, there are certain
such scenarios, cross-matching becomes feasible for tracking issues which are not addressed by these systems. First, though
individual users [27][30][31]. these systems secure cryptographic key using biometric tem-
2) Biometric Key Release: These are the systems where plate, the template itself still remains unprotected. This leads to
biometrics along with cryptographic keys are used for au- all security and privacy issues discussed earlier. Second, these
thentication and communication [32]. The effort lies in using systems fail to provide revocability of biometric templates
biometric templates effectively to release cryptographic keys meaning that if it is known that biometric template of a

Copyright (c) IARIA, 2012. ISBN: 978-1-61208-221-9 24

PATTERNS 2012 : The Fourth International Conferences on Pervasive Patterns and Applications

particular user has been compromised, it is not feasible for accuracy of iris recognition systems. These are detailed as
him/her to change his/her secret in contrast to password or follows.
hardware token-based systems. 1) Dilation: One of the important but often ignored factor
3) Cancelable Biometrics: Cancelable biometric systems is pupil dilation. Due to dilation effects, we have varying size
apply some transformation on the biometric template to secure of pupil, which results in decreased recognition performance.
the template [6][21][33][42]. The idea is that, instead of Dilation may occur due to many factors such as drugs,
directly storing the template, a function is applied on the sunglasses, light illumination, etc.
template and the output of that function (transformed template) Experimental studies are presented by Hollingsworth et al.
is stored in the database. The transformation function must identifying the effects of pupil dilation on iris recognition
be non-invertible so that a compromised transformed template performance [15][16]. To produce dilation, they used ambient
cannot be translated to the original template. The major light for controlled intervals of time. Degree of dilation was
advantage of cancelable biometric systems is that even if the measured by taking the fraction of pupil and iris radius. They
transformed template is compromised, the original template conducted two experiments, one to find out the effects of
still remains secure. In addition, the secret can easily be dilation of same degree (between two templates to be matched)
revoked by applying a different transformation to the original and second with varying dilation. Their findings indicate that
template resulting in a new transformed template. Moreover, 1) If both images have same but high pupil dilation, this results
a user can have different transformed templates for different in lower recognition performance as compared to images
applications he/she is using hence making cross-matching with no dilation. 2) If images have different amount of pupil
infeasible for any potential attacker. dilation, this results in further increasing of False Reject Rate
Finding a suitable transformation function can be quite (FRR).
tricky in cancelable biometric systems. Standard non-invertible Effects of pupil dilation on iris recognition performance
transformation functions (such as one-way cryptographic hash have been studied by other researchers also [4][10][29]. Rakshit
functions) do not work with biometric data due to intra- and Monro [29] have used eye drops to achieve the effects of
class variability of biometric data. Therefore, in most cases, dilation. For their experiments, they collected images before
transformation is user-dependent, i.e., user either has to re- and after 5, 10 and 15 minutes of instilling of drops. In
member a password/pin or to carry a token which stores most cases, due to the instillation of drops, pupil lost its
the transformation parameters. This puts an extra burden on shape and they used their shape-description method to generate
the user and effectively converts the system into two-factor accurate normalized images. Their experiments also showed
authentication scheme. It is also desirable to observe user a decrease in recognition performance due to iris dilation.
specific key to check the strength of user-provided secret. They also observed that, with the increase in time, dilation
4) Biometric Key Generation: In such systems, bio- is increased leading to an increase in FRR. Dhir et al. [10]
metric template and cryptographic key are bind together later extended their study with 15 subjects as compared to 11
[11][14][20][22][26][41]. Cryptographic key can either be in [29]. They found the same results namely dilation results in
generated directly from biometric template [14][20][22][41] poor performance and false reject rate increases with increase
or by using standard cryptographic techniques [11][26]. In in dilation which in turn increases with time after eye drops
former case, generated key is not uniform and hence may not have been administered.
be strong enough for use in many cryptographic protocols. In Bowyer et al. [4] categorized iris images in three classes
biometric key generation systems, neither biometric template based on amount of pupil dilation namely small, medium and
nor cryptographic key is stored in cleartext. Instead, a value large. For experiments with varying amount of dilation, their
obtained by binding these two secrets is stored such that it results show that the larger the difference in dilation ratio, the
is not feasible to get any of the two secrets from this bound more the chances of false non-match. For experiments with
value. same amount of dilation, their findings indicate that increasing
Though last three non-traditional systems described above the degree of dilation, increases the false match and false non-
are quite effective in resolving template security related issues match.
in biometric recognition systems, in most cases, recognition From the above studies, it can safely be concluded that it
performance is affected. In addition, speed of these systems is not that difficult to deceive iris recognition systems which
is always slower as compared to conventional iris recognition is contrary to the popular belief in research community. Pupil
systems. Moreover, most of these systems do not perform well dilation not only affects the recognition performance but an
with noisy iris image datasets. Due to all these issues, we can intruder can easily deceive the system by just wearing sun-
conclude that a reliable and efficient solution to solve template glasses or by using eye drops. Pupil dilation factor should be
security related issues is yet to be achieved. incorporated in iris recognition systems to increase confidence
in recognition results.
B. Iris Recognition Performance 2) Lenses: Around the world, approximately 125 million
An iris recognition system is considered ideal when match people use contact lenses. Therefore, iris recognition systems
and non-match distributions do not overlap each other. There should be flexible enough to accommodate these large number
are a few factors which may lead to a significant drop in of people. Designers of iris recognition algorithms claim that

Copyright (c) IARIA, 2012. ISBN: 978-1-61208-221-9 25

PATTERNS 2012 : The Fourth International Conferences on Pervasive Patterns and Applications

recognition performance of their systems is not affected by similarities between irises of twins.
the use of contact lenses[1][8][28][40]. But, recently, Baker et They conducted two user studies. First, where only irises
al. [2] come up with a study showing that every type of lens of subjects are presented to respond to the queries and second
negatively affects iris recognition performance. They used a where periocular images are displayed to the user to respond
dataset containing 51 subjects with contact lens and 64 without to the queries. On the iris image experiments, the average
lenses. After visual inspection of iris images, they categorized success score is 81.3% and for periocular queries success
lenses into four categories. First category includes lenses that score is 76.5%. Their findings indicate that there are simi-
are visible but have no effects on the iris. Second category larities between the irises of genetically same users which can
includes images that result in light or dark outline around iris be visually identified, but current biometric systems do not
and sclera. Third category includes lenses with large artefacts identify them. It is required to explore further and establish
on the iris that are mainly due to written logo/number or misfit techniques so that biometric systems may utilize this visual
lens. Fourth category is one having subjects with hard lenses. similarity between genetically similar irises for the benefit of
They conducted two experiments. First experiment com- performance enhancement.
pares results of contact lenses and non-contact lenses subjects 4) Time Variability: Human iris is considered stable over
while the second experiment compares results of different time [23][24][25][34]; but, a recent study by Gonzalez et al. [36]
categories of contact lenses. In first experiment, false reject shows results which contradict what has been demonstrated
rate for subjects with lenses is 9.42% and 0.719% for subjects so far. For their experimental evaluation, Gonzalez et al. used
without contact lenses. This shows that contact lenses have BioSecurId [12] and BioSec [13] baseline datasets. The former
a severely adverse effect on iris recognition accuracy. The dataset consist of 254 individuals (8128 images) captured
second experiment showed that second category is the one with in four different sessions and later has 200 subjects (3200
the lowest FRR of 3.9% whereas fourth category has highest images) captured in two different sessions, both splitted by a
FRR of 45.44%. Category one and three have also shown high time span of one to four weeks. Results show that errors rate
false reject rates of 10.64% and 14.37%, respectively. As is is increased in inter-session experiments compared with the
clear from results, lenses of all types affect the verification intra-session ones. Their finding indicates that, as the lapse
results little or more depending on the type. time between enrolment and comparison is increased, false
Baker et al. [3] later conducted a larger study on the accept rate remains unaffected but false reject rate is increased
effects of lenses. They used three different systems for iris up to more than twice. Bowyer et al. [4] conducted a similar
recognition namely IrisBEE, VeriEye and CMU. They also research to find the effect of time variability on recognition
categorised lenses in four types. The results show that false performance. Their recognition results also showed that as the
reject rate for subjects with lenses is much higher than that time between enrolment and verification increases, false reject
for subject without lenses. In addition, category four of hard rate of the system also increases though that increase is less
lenses showed worst recognition results among all lens types significant than that reported by Gonzalez et al. Although,
for all three iris recognition systems. research results show that time variability affects verification
Bowyer et al. [4] conducted a similar study to evaluate performance but to be more confident in extent of this effect,
iris recognition performance among subjects wearing contact more research with large datasets is desirable.
lenses. Their findings are that false non-match score was 5) Cataract Surgery: In [10] Dhir et al. and [29] Rakshit
almost same for contact lens and non-contact lens groups while et al. identified the effects of cataract surgery on recognition
false match score was 0.27% for non-contact lens group and performance. In [29], they collected the images of 3 patients
5.64% for contact lens group showing a significant drop in before and after two weeks of cataract surgery. The results of
recognition accuracy. From the above reported studies and pre and post surgery images comparison shows that cataract
results, the effects of the contact lenses are apparent on recog- surgery does not affect recognition performance. Later on,
nition performance. All types of lenses result in performance Dhir et al. [10] did similar experiments with 15 subjects and
degradation so there is need to introduce techniques that can found same results. Although the study is significant, but as
handle such scenarios to strengthen iris biometric systems. the dataset was not large, there is need to do more experiments
3) Twins: In [17][18], Hollingworth et. al presented studies on larger datasets to explore the effects.
identifying the texture similarities between irises of twins. 6) System Portability: To check system portability related
Their work is in contrast to the previous work which focuses issues, Bowyer et al. [4] performed experiments on a set of iris
on identifying the differences between genetically same iden- images acquired using different sensors namely LG 2200 and
tities. To conduct their experiments, they collected the data on LG 4400. Experiments show false reject rate is higher when
twins day festival Twinburg in Ohio in August 2009. both images (enrolment and verification) are from different
They also collected the data from unrelated people to do sensors compared with the results where both images are from
comparative analysis. At first step, they performed biometric the same sensor. The study is done on limited dataset and only
system testing and their findings are same as those of the old using IrisBEE software. There are chances that results may
researchers, i.e., for iris biometric system, irises of twins are be affected differently by different software and hardware. A
more or like similar as those belonging to unrelated people. larger research is needed to explore effect of different sensors
At the second step, they performed user testing to identify on iris recognition performance.

Copyright (c) IARIA, 2012. ISBN: 978-1-61208-221-9 26

PATTERNS 2012 : The Fourth International Conferences on Pervasive Patterns and Applications

IV. C ONCLUSION AND F UTURE W ORK [18] K. Hollingsworth, K. Bowyer, S. Lagree, S. Fenker, and P. Flynn.
Genetically identical irises have texture similarity that is not detected
This review paper summarizes the issues and challenges by iris biometrics. Computer Vision and Image Understanding, pages
with current iris biometric systems. In particular, we discussed 14931502, 2011.
[19] M. Hosseini, B. Araabi, and H. Soltanian-Zadeh. Pigment melanin:
security and performance related issues. We have shown that pattern for iris recognition. IEEE Transactions on Instrumentation and
many popular beliefs about security, reliability, stability and Measurement, 59(4):792804, 2010.
performance of iris recognition systems are not correct and [20] S. Kanade, D. Camara, E. Krichen, D. Petrovska-Delacretaz, and
B. Dorizzi. Three factor scheme for biometric-based cryptographic key
need to be revisited. The issues raised in this survey should regeneration using iris. In Biometrics Symposium, 2008. BSYM08, pages
be addressed in order to be more confident in working of 5964. IEEE, 2008.
iris recognition systems. In the future, we plan to explore the [21] S. Kanade, D. Petrovska-Delacretaz, and B. Dorizzi. Cancelable iris
security and privacy concerns facing other biometric systems. biometrics and using error correcting codes to reduce variability in
biometric data. In Computer Vision and Pattern Recognition, 2009.
This could lead to a design of multi-biometric system that CVPR 2009. IEEE Conference on, pages 120127. IEEE, 2009.
overcomes the weaknesses of one by the strength of other [22] Y. Lee, K. Bae, S. Lee, K. Park, and J. Kim. Biometric key binding:
biometric. Fuzzy vault based on iris images. In 2nd International Conference on
Biometrics, pages 800808. Springer, 2007.
[23] I. Maghiros, Y. Punie, S. Delaitre, E. Lignos, C. Rodriguez, M. Ulbrich,
R EFERENCES and M. Cabrera. Biometrics at the frontiers: Assessing the impact on
society. Institute for Prospective Technological Studies, Technical Report
[1] J. Ali and A. Hassanien. An iris recognition system to enhance e-security EUR, 21585, 2005.
environment based on wavelet theory. AMO-Advanced Modeling and [24] K. Miyazawa, K. Ito, T. Aoki, K. Kobayashi, and H. Nakajima. An
Optimization, 5(2):93104, 2003. effective approach for iris recognition using phase-based image match-
[2] S. Baker, A. Hentz, K. Bowyer, and P. Flynn. Contact lenses: Handle ing. IEEE Transactions on Pattern Analysis and Machine Intelligence,
with care for iris recognition. In International Conference on Biometrics: 30(10):17411756, 2008.
Theory, Applications, and Systems (BTAS), pages 18, 2009.
[25] D. Monro, S. Rakshit, and D. Zhang. DCT-based iris recognition. IEEE
[3] S. Baker, A. Hentz, K. Bowyer, and P. Flynn. Degradation of iris Transactions on Pattern Analysis and Machine Intelligence, 29(4):586
recognition performance due to non-cosmetic prescription contact lenses. 595, 2007.
Computer Vision and Image Understanding, 114:1030044, 2010.
[26] F. Monrose, M. Reiter, Q. Li, and S. Wetzel. Cryptographic key
[4] K. Bowyer, S. Baker, A. Hentz, K. Hollingsworth, T. Peters, and
generation from voice. In Security and Privacy, 2001. S&P 2001.
P. Flynn. Factors that degrade the match distribution in iris biometrics.
Proceedings. 2001 IEEE Symposium on, pages 202213. IEEE, 2001.
Identity in the Information Society, 2(3):327343, 2009.
[27] K. Nandakumar, A. Jain, and S. Pankanti. Fingerprint-based fuzzy vault:
[5] K. Bowyer, K. Hollingsworth, and P. Flynn. Image understanding for
Implementation and performance. IEEE Transactions on Information
iris biometrics: A survey. Computer Vision and Image Understanding,
Forensics and Security, 2(4):744757, 2007.
110(2):281307, 2008.
[28] M. Negin, T. Chmielewski Jr, M. Salganicoff, U. von Seelen, P. Vene-
[6] J. Bringer, H. Chabanne, and B. Kindarji. The best of both worlds:
tainer, and G. Zhang. An iris biometric system for public and personal
Applying secure sketches to cancelable biometrics. Science of Computer
use. Computer, 33(2):7075, 2000.
Programming, 74(1-2):4351, 2008.
[7] J. Daugman. High confidence visual recognition of persons by a test [29] S. Rakshit and D. Monro. Medical conditions: Effect on iris recogni-
of statistical independence. IEEE Transactions on Pattern Analysis and tion. In Multimedia Signal Processing, 2007. MMSP 2007. IEEE 9th
Machine Intelligence, 15(11):11481161, 1993. Workshop on, pages 357360. IEEE, 2007.
[8] J. Daugman. How iris recognition works. IEEE Transactions on Circuits [30] N. Ratha, S. Chikkerur, J. Connell, and R. Bolle. Generating cancelable
and Systems for Video Technology, 14(1):2130, 2004. fingerprint templates. Pattern Analysis and Machine Intelligence, IEEE
[9] S. Dhavale. Robust iris recognition based on statistical properties of Transactions on, 29(4):561572, April 2007.
walsh hadamard transform domain. International Journal of Computer [31] N. Ratha, J. Connell, R. Bolle, and S. Chikkerur. Cancelable biometrics:
Science, 9, 2012. A case study in fingerprints. In Pattern Recognition, 2006. ICPR 2006.
[10] L. Dhir, N. Habib, D. Monro, and S. Rakshit. Effect of cataract surgery 18th International Conference on, volume 4, pages 370373, 2006.
and pupil dilation on iris pattern recognition for personal authentication. [32] O. Song, A. Teoh, and D. Ngo. Application-specific key release scheme
Eye, 24(6):10061010, 2009. from biometrics. International Journal of Network Security, 6(2):127
[11] H. Feng and C. Wah. Private key generation from on-line handwritten 133, 2008.
signatures. Information Management & Computer Security, 10(4):159 [33] A. Teoh and C. Yuang. Cancelable biometrics realization with mul-
164, 2002. tispace random projections. Systems, Man, and Cybernetics, Part B:
[12] J. Fierrez, J. Galbally, J. Ortega-Garcia, M. Freire, F. Alonso-Fernandez, Cybernetics, IEEE Transactions on, 37(5):10961106, 2007.
D. Ramos, D. Toledano, J. Gonzalez-Rodriguez, J. Siguenza, J. Garrido- [34] J. Thornton, M. Savvides, and V. Kumar. A Bayesian approach to
Salas, et al. BiosecurID: a multimodal biometric database. Pattern deformed pattern matching of iris images. IEEE Transactions on Pattern
Analysis & Applications, 13(2):235246, 2010. Analysis and Machine Intelligence, 29(4):596606, 2007.
[13] J. Fierrez, J. Ortega-Garcia, D. Toledano, and J. Gonzalez-Rodriguez. [35] J. Thornton, M. Savvides, and B. Vijayakumar. Robust iris recognition
Biosec baseline corpus: A multimodal biometric database. Pattern using advanced correlation techniques. Image Analysis and Recognition,
Recognition, 40(4):13891392, 2007. pages 10981105, 2005.
[14] F. Hao, R. Anderson, and J. Daugman. Combining cryptography with [36] P. Tome-Gonzalez, F. Alonso-Fernandez, and J. Ortega-Garcia. On the
biometrics effectively. University of Cambridge Computer Laboratory, effects of time variability in iris recognition. In IEEE International
Tech. Rep, 2005. Conference on Biometrics: Theory, Applications and Systems, (BTAS),
[15] K. Hollingsworth, K. Bowyer, and P. Flynn. The importance of small pages 16. IEEE, 2008.
pupils: a study of how pupil dilation affects iris biometrics. In 2nd [37] Z. Wang, Q. Han, and C. Busch. A novel iris location algorithm.
IEEE International Conference on Biometrics: Theory, Applications and International Journal of Pattern Recognitionand Artificial Intelligence,
Systems, 2008. BTAS 2008., pages 16. IEEE, 2008. 23(1):59, 2009.
[16] K. Hollingsworth, K. Bowyer, and P. Flynn. Pupil dilation degrades iris [38] R. Wildes. Iris recognition: An emerging biometric technology. PIEEE,
biometric performance. Computer Vision and Image Understanding, 85(9):13481363, September 1997.
113(1):150157, 2009. [39] R. Wildes, J. Asmuth, G. Green, S. Hsu, R. Kolczynski, J. Matey, and
[17] K. Hollingsworth, K. Bowyer, and P. Flynn. Similarity of iris texture S. McBride. A machine-vision system for iris recognition. Machine
between identical twins. In IEEE Computer Society Conference on Vision and Applications, 9(1):18, 1996.
Computer Vision and Pattern Recognition Workshops (CVPRW), pages [40] G. Williams. Iris recognition technology. IEEE Aerospace and Elec-
2229. IEEE, 2010. tronic Systems Magazine, 12(4):2329, 1997.

Copyright (c) IARIA, 2012. ISBN: 978-1-61208-221-9 27

PATTERNS 2012 : The Fourth International Conferences on Pervasive Patterns and Applications

[41] S. Ziauddin and M. Dailey. Robust iris verification for key management.
Pattern Recognition Letters, 31(9):926935, 2010.
[42] J. Zuo, N. Ratha, and J. Connell. Cancelable iris biometric. In Pattern
Recognition, 2008. ICPR 2008. 19th International Conference on, pages
14. IEEE, 2008.

Copyright (c) IARIA, 2012. ISBN: 978-1-61208-221-9 28