Documente Academic
Documente Profesional
Documente Cultură
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/271430750
CITATIONS READS
12 189
5 authors, including:
Some of the authors of this publication are also working on these related projects:
Secure Access Control over Wide Area Network (SWAN) (248030/O70) View project
All content following this page was uploaded by R. Raghavendra on 17 August 2015.
Kiran B. Raja, R. Raghavendra, Faouzi Alaya Cheikh, Bian Yang, Christoph Busch
ABSTRACT and even for identical twins. Biometric features such as face
Iris is one of the preferred biometric modalities. Never- or fingerprint are always at the risk of being changed due to
theless, the focus of iris image has to be good enough various factors. The performance of the recognition system
to achieve good recognition performance. Traditional iris depends largely on the change in facial expressions based on
imaging devices in the visible spectrum suffer from limited the social factors and also throws in challenges in recognition
depth-of-field which results in out-of-focus iris images. The with varied illumination, age and pose [3]. Another well
acquisition of iris image is thus repeated until a satisfactory known biometric modality is fingerprint. Any intentional or
focus is obtained or the image is post-processed to improve unintentional scars or cuts on the fingerprint may introduce
the visibility of texture pattern. Bad focused images obtained false recognition or rejection of authentic subjects. Fake
due to non-optimal focus degrade the identification rate. fingerprint attack has to be considered for a secure biometric
In this work, we propose a novel scheme to capture high recognition. These factors influence the intra-class variability
quality iris samples by exploring new sensors based on to larger extent and thereby make the inter-class variability
light-field technology to address the limited depth-of-field lower. Lower inter-class variability leads to challenges in
exhibited by the conventional iris sensors. The idea stems accurate recognition. Thus, iris provides two unique bio-
out from the availability of multiple depth/focus images metric identities for any single person with a high level of
in a single exposure. We propose to use the best-focused identification confidence. Human iris is also one of the most
iris image from the set of depth images rendered by the distinctive features for each individual and is thus used for
Light-field Camera (LFC). We further evaluate the proposed robust biometric recognition. Owing to the robust level of
scheme experimentally with a unique and newly acquired identity protection it provides, iris recognition is unparalleled
iris database simulating the real-life scenario. by any other biometric feature. Iris biometric feature is not
prone to the vast changes or morphing over the period of
Index Terms— Biometrics, Iris recognition, Light-field
lifetime.
camera, out-of-focus iris.
To capture the iris pattern presented by the dark pigmented
I. INTRODUCTION iris, traditional methods have employed the Near-Infra-Red
The human iris is defined as a thin circular diaphragm illumination in the range of 700 − 900 nm. Atos Origin
lying between the cornea and the lens in the eye [1]. Iris [4] has reported that stop-and-stare strategy followed by
is one of the organs present internally in human body but many iris recognition systems to acquire good quality images
also visible externally when the eye-lids are open. The iris in the Near-Infra-Red region remains a major hurdle in
is known to develop in the third month of gestation and deploying the iris-based biometrics in large scale for faster
prominent structures resulting in the patterns are mostly identification/recognition. Many of the currently deployed
complete by eighth month [18]. It contains rich amount of systems impose constraints on the subjects and acquisition
texture and unique structures like furrows, freckles, crypts, environment. A subject is expected to stand close to the
and coronas [1]. Color of the iris is known to vary indi- imaging device and co-operatively look into the camera
vidually for each person. The color is related to density of under NIR illumination to assure good quality images. To
melanin pigment in the anterior layer and stroma [19] in iris. achieve a good signal-to-noise ratio in the sensor and to
The presence of lower amount of melanin pigment in iris capture images with highly discriminating iris features, high
results in light colored iris and the higher amount results in illumination is required [5]. This becomes non-acceptable
darker iris. Light colored iris allows the penetration of long- for acquisitions under at-a-distance or on-the-move scenarios
wavelength light and usually scatters shorter wavelength because of the amount of light required for the process. To
light [2]. One important aspect to consider is the epigenetic broaden the scope of iris recognition in real-life at-a-distance
nature of iris patterns. This results in unique, completely and on-the-move surveillance scenarios, we have focused our
independent and uncorrelated iris patterns for an individual work on visible spectrum iris recognition.
Light Field Different Iris ROI Pre-Processing Energy Based III-B. Best Focused Iris Image Selection
Camera Dept h (Focus) Select ion Yes
Feat ure Subject
Verificat ion
Compare Ident ified The main motivation of this work is to obtain the best
Best Focused
focused image after capture. Of all the multiple depth-
Segmentat ion
Normalized Iris
Feat ure images obtained, the image having the highest energy is
considered for the recognition. Better focus for image cap-
Normalizat ion Ext ract ion
Wavelet Energy
3.67
results in four sub-images ILL (x, y), ILH (x, y), IHL (x, y)
and IHH (x, y) where, ILL (x, y) is smoothed image corre- 3.00
sponding to the low-frequency band that represent the coarse
approximation of the original image I(x, y) and termed 2.00 1.68 1.77
as AI . ILH (x, y), IHL (x, y) and IHH (x, y) represents the 1.29 1.18
The best-focused iris image Ibf (x, y) is used for feature IV. EXPERIMENTS AND RESULTS
extraction. The quantitative results of the experiments on our newly
Figure 5 shows the measured focus in-terms of the wavelet collected Iris dataset using both Light-field and conventional
energy for iris samples acquired for a single subject using camera are discussed in this section. A total of 84 unique iris
light-field and conventional cameras. The best-focused light- data obtained each from Light-field camera and conventional
field images have higher energy and thus higher focus 2D camera were used in this work. The experiments were
compared to the conventional iris images. Figure 6 shows the conducted using 420 iris images corresponding to conven-
energy measured for various depth images for the normalized tional camera and 420 best focused images corresponding to
iris captured from light-field camera in single exposure. The light-field camera. 3387 depth images were used to measure
depth represented by D 4 gives highest energy and thus is the focus to obtain 420 best focused images. The whole set
chosen to carry out the performance evaluation. of iris images was evaluated for the identification accuracy
using the state-of-the-art iris recognition algorithms [19], 100
1 − FNMR (%)
equals False Non-Match Rate (FNMR) [25]. Table I shows 50
the performance of the best-focus images obtained by the 20 Rathgeb & Uhl [21]
Rathgeb & Uhl [23]
light-field camera exceeds by 4.64% on average as compared 10
Ko et al. [24]
Ma et al. [22]
to the conventional camera and the best performance with Masek et al. [20]
Daugman [19]
Table I. Quantitative results obtained from various schemes Fig. 8. Receiving Operator Characteristic graph for best-
EER (%) focused iris images from light-field camera.
Feature Extraction
Light-field Conventional
Camera camera
Rathgeb & Uhl [21] 26.92 27.38
V. CONCLUSIONS
Rathgeb & Uhl [23] 11.84 18.24
Ko et al. [24] 5.95 12.55
Ma et al. [22] 5.78 12.07
The main motivation of this work is to address the out-
Masek et al. [20] 5.26 12.44
of-focus images in the conventional iris imaging systems
Daugman [19] 2.38 8.53 under visible spectrum. We have explored the strength of
the light-field technology for the benefit of iris imaging.
Receiver operating characteristic (ROC) graphs for light- Multiple depth images obtained from single exposure are
field camera and conventional camera are presented in Figure analyzed using the newly proposed energy based measure
7 and 8 respectively. The ROC shows the clear increase in to obtain a best focused image. The best focused image is
1 − F N M R at F M R = 10−1 for the light-field camera then employed for iris recognition to address the draw-backs
in Figure 8. Thus from the results given in this section, it of out-of-focus iris images. Due to limited depth-of-field
has been experimentally verified that the light-field camera in conventional cameras for iris imaging the acquisition of
provides much higher accuracy in the visible spectrum iris iris samples has to be repeated to obtain the best-focus. We
recognition. have employed the Lytro light-field camera to experimentally
100
validate our proposed scheme. The whole experiment was
conducted on 84 unique irises with 5 samples for each iris
90
captured using conventional camera and light-field camera.
80 This work has resulted in the creation of a unique iris
70
database captured with conventional camera and light-field
camera with 420 sample images along with 3387 multiple-
60
focus images. This light-field iris image database is the first
1 − FNMR (%)
50
and unique database to the best of our knowledge. The
40 outcomes of the experiment have shown the success of the
30
proposed scheme. The comparison of the performance of
best-focused light-field iris images with conventional iris
Rathgeb & Uhl [21]
20
Rathgeb & Uhl [23] images has indicated better performance for the proposed
Ko et al. [24]
10 Ma et al. [22]
Masek et al. [20]
scheme. Best-focused images obtained from the light-field
0 −1 0 1
Daugman [19]
2
camera reported an EER of 2.38% with 2D Gabor features
10 10
FMR (%)
10 10
proposed by Daugman [19] (refer to Table I), while the
degraded performance has been demonstrated of 26.92% for
Fig. 7. Receiving Operator Characteristic graph for conven- the context based iris features [21]. The light-field images
tional iris images. have given an average improvement of around 4.64% as
compared to conventional images.
VI. ACKNOWLEDGMENT [12] Lytro Inc, ,” http://www.lytro.com/.
This work was supported by the Erasmus Mundus Master [13] Ren Ng, Marc Levoy, Mathieu Brédif, Gene Duval,
CIMET and EU 7th Framework Program (FP7/2007-2013) Mark Horowitz, and Pat Hanrahan, “Light field photog-
under grant agreement no 284862 for the large-scale inte- raphy with a hand-held plenoptic camera,” Computer
grated project FIDELITY. Science Technical Report CSTR, vol. 2, 2005.
[14] Ashok Veeraraghavan, Ramesh Raskar, Amit Agrawal,
VII. REFERENCES Ankit Mohan, and Jack Tumblin, “Dappled photog-
[1] FH Adler, “Physiology of the eye, mosby, st,” Louis, raphy: Mask enhanced cameras for heterodyned light
Mo, 1965. fields and coded aperture refocusing,” ACM Transac-
[2] MR Chedekel, “Photophysics and photochemistry of tions on Graphics, vol. 26, no. 3, pp. 69, 2007.
melanin,” Melanin: Its Role in Human Photoprotection, [15] R. Raghavendra, Bian Yang, Kiran B. Raja, and
pp. 11–23, 1995. Christoph Busch, “A new perspective - face recognition
[3] Yael Adini, Yael Moses, and Shimon Ullman, “Face with light-field camera,” in 6th IAPR International
recognition: The problem of compensating for changes Conference on Biometrics (ICB), Madrid. IEEE, 2013.
in illumination direction,” Pattern Analysis and Ma- [16] Nobuyuki Otsu, “A threshold selection method from
chine Intelligence, IEEE Transactions on, vol. 19, no. gray-level histograms,” Automatica, vol. 11, no. 285-
7, pp. 721–732, 1997. 296, pp. 23–27, 1975.
[4] Atos Origin, “Uk passport service biometrics enrolment [17] Guillaume Sutra, Bernadette Dorizzi, Sonia Garcia-
trial,” Report. Retrieved April, vol. 15, pp. 2009, 2005. Salicetti, and Nadia Othman, “A biometric reference
[5] Hugo Proença, “Iris recognition in the visible wave- system for iris, osiris version 4.1,” 2012.
length,” in Handbook of Iris Recognition, pp. 151–169. [18] Thierry Lefevre, Bernadette Dorizzi, Sonia Garcia-
Springer, 2013. Salicetti, Nadege Lemperiere, and Stephane Belardi,
[6] Kevin W Bowyer, Karen Hollingsworth, and Patrick J “Effective elliptic fitting for iris normalization,” Com-
Flynn, “Image understanding for iris biometrics: A puter Vision and Image Understanding, 2013.
survey,” Computer vision and image understanding, [19] John Daugman, “How iris recognition works,” in Im-
vol. 110, no. 2, pp. 281–307, 2008. age Processing. 2002. Proceedings. 2002 International
[7] Nathan D Kalka, Jinyu Zuo, Natalia A Schmid, and Conference on. IEEE, 2002, vol. 1, pp. I–33.
Bojan Cukic, “Estimating and fusing quality factors for [20] Libor Masek, “Recognition of human iris patterns for
iris biometric images,” Systems, Man and Cybernetics, biometric identification,” M.S. thesis, University of
Part A: Systems and Humans, IEEE Transactions on, Western Australia, 2003.
vol. 40, no. 3, pp. 509–524, 2010. [21] Christian Rathgeb and Andreas Uhl, “Context-based
[8] Ramkumar Narayanswamy, Paulo EX Silveira, Harsha texture analysis for secure revocable iris-biometric key
Setty, V Paul Pauca, and Joseph van der Gracht, generation,” in Crime Detection and Prevention (ICDP
“Extended depth-of-field iris recognition system for a 2009), 3rd International Conference on. IET, 2009, pp.
workstation environment,” in Defense and Security. 1–6.
International Society for Optics and Photonics, 2005, [22] Li Ma, Tieniu Tan, Yunhong Wang, and Dexin Zhang,
pp. 41–50. “Personal identification based on iris texture analy-
[9] Naresh Boddeti and BVKV Kumar, “Extended depth sis,” Pattern Analysis and Machine Intelligence, IEEE
of field iris recognition with correlation filters,” in Transactions on, vol. 25, no. 12, pp. 1519–1533, 2003.
Biometrics: Theory, Applications and Systems, 2008. [23] Christian Rathgeb and Andreas Uhl, “Secure iris recog-
BTAS 2008. 2nd IEEE International Conference on. nition based on local intensity variations,” in Image
IEEE, 2008, pp. 1–8. Analysis and Recognition, pp. 266–275. Springer, 2010.
[10] Kang Ryoung Park and Jaihie Kim, “A real-time [24] Jong-Gook Ko, Youn-Hee Gil, Jang-Hee Yoo, and Kyo-
focusing algorithm for iris recognition camera,” Sys- IL Chung, “A novel and efficient feature extraction
tems, Man, and Cybernetics, Part C: Applications and method for iris recognition,” ETRI journal, vol. 29,
Reviews, IEEE Transactions on, vol. 35, no. 3, pp. 441– no. 3, pp. 399–401, 2007.
444, 2005. [25] ISO/IEC TC JTC1 SC37 Biometrics, ISO/IEC 19795-
[11] Sriram Pavan Tankasala, Vikas Gottemukkula, 1:2006. Information Technology – Biometric Perfor-
Sashi Kanth Saripalle, Venkata Goutam Nalamati, mance Testing and Reporting – Part 1: Principles and
Reza Derakhshani, Raghunandan Pasula, and Arun Framework, International Organization for Standard-
Ross, “A video-based hyper-focal imaging method ization and International Electrotechnical Committee,
for iris recognition in the visible spectrum,” in Mar. 2006.
Homeland Security (HST), 2012 IEEE Conference on
Technologies for. IEEE, 2012, pp. 214–219.