Li S.Z., Jain A.K. (eds.) Encyclopedia of Biometrics

Подождите немного. Документ загружается.

Iris Capture

▶ Biometric Sample Acquisition

Iris Data Interchange Standards

▶ Iris Image Data Interchange Formats ,

Standardization

Iris Databases

DAM ON L. WOODAR D

,KARL RICANEK

Clemson University, Clemson, SC, USA

University of North Carolina Wilmington,

Wilmington, NC, USA

Definition

An iris database is a collection of images that contain,

at a minimum, the iris region of the eye. The images

are typically collected by sensors that op erate in

the

▶ visible spectrum, 380–750 nm, or the near infra-

red spectrum (NIR), 700–900 nm. The visible spec-

trum image can be stored as a color image or as an

intensity image. The NIR image is always stored as an

intensity image.

Introduction

Successful biometric research requires the analysis of

human data. For biometric researchers to demonstrate

the effectiveness of proposed iris segmentation/

recognition techniques and allow fair comparisons

with existing methods, publicly available iris databases

are required. The perfect iris-image database should be

sufﬁciently large, consist of images collected from a

large and heterogeneous group of subjects, and contain

images that depict noise factors typically encountered

in real world applications. In the following sections,

several publicly and freely available iris-image data-

bases are described.

CASIA V1.0, V2.0, IrisV3 Databases

The CASIA Iris Image databases were collected by the

Institute of Automation at the Chinese Academy of

Sciences. The original CASIA database (CASIA V1.0)

is one of the oldest publicly available systems for eval-

uation of the iris biometric modality; hence, CASIA

V1.0 has been widely used for research and evaluation.

The database [1] consists of 756 320  280 intensity

iris images of 108 eyes captured using a sensor devel-

oped in-house. The images are stored as 8 bit gray-level

JPEG ﬁles. Seven images are captured from each eye

during two sessions, three during the ﬁrst session and

four during the second. The data acquisition environ-

ment was hig hly constrained which limited the noise

types present in the image to iris occlusion from eye-

lids and eye lashes. It was determined that the iris

images of the CASIA V1.0 database had been photo-

graphically edited by replacing the pupil region with a

circular region of uniform intensity [2], which is illu-

strated in Fig. 1. Therefore, this version of CASIA

should not be used for algorithm development or

evaluation.

The CASIA V2.0 database was released later and

contains 2,400 images collected from 60 eyes. Twenty

images per eye were collected using two different sen-

sors, the in-house sensor and the OKI Iris Pass system.

In order to receive copy of CASIA V2.0, CASIA V1.0

must be downloaded ﬁrst.

The CASIA-IrisV3 datab ase consists of 22,051

images captured from 1,500 eyes of more than 700

subjects, organized three separate databases of dis-

joint subject groups [3]. The ‘‘Interval’’ subset contains

2,655 images captured from 396 eyes of 249 subjects.

The 320  280 intensity images were captured indoors

using a self-developed sensor that was used in CASIA

V1.0 during two acquisition sessions over at least a

Iris Databases. Figure 1 Example images from CASIA

V1.0 and CASIA-IrisV3 ‘‘Interval’’ databases. (a) An image

from CASIA v1.0 (b) An image from CASIA-IrisV3.

770

Iris Capture

1 month time. Included in the ‘‘Interval’’ database is an

unedited – the pupil region was not replaced to mask

the NIR illumination pattern – sup erset of the CASIA

V1.0 database. An example of the unedited image can

be found in Fig. 1(b).

The ‘‘Lamp’’ database contains 16,213 images cap-

tured from 819 eyes of 411 subjects. The 640  480

grayscale images were captured indoors with varying

illumination with the OKI IRISPASS-h sensor during a

single acquisition session. The ‘‘Twins’’ database con-

tains 3,183 imag es captured from 400 eyes of 200

subjects. The 640  480 intensity images were captured

outdoors using the OKI IRI SPASS-h sensor during a

single acquisition session.

UPOL Database

The UP OL iris-image database was constructed at the

University of Palacke

ho and Olomouc in the Czech

Republic [4, 5]. The database contains 384 768  576

color images captured from 128 eyes of 64 subjects

(three images per left and right eye). The images are

stored as 24-bit color images in the PNG image format.

Unlike the other iris-image databases, the UPOL data-

base was acquired using an optometric sensor (TOP-

CON TRC50IA optical device connected to a SONY

DXC-950P camera). As a result, the images are of

very high quality and are practically noiseless, as

shown in Fig. 2.

BATH Database

The University of Bath, UK, iris-image database is

composed mainly of images captured from its ethni-

cally diverse staff and students [6]. The database

contains over 16,000 1280  960 iris images collected

from each eye of 800 subjects. A commercial version of

this database claims to be twice as large. The sensor

used for data collection was the ISG LightWise LW-1.3-

S-1394. Illumination was provided using an array of

infrared LEDs positioned such that reﬂections were

conﬁned to the pupil area w ithin the image, which is

illustrated in Fig. 2. An infrared pass ﬁlter was used to

reduce environmental light reﬂections. The researchers

were able to achieve high qualit y images using this

framework. The main sources of noise within the

images are due to the occlusion of the iris by eyelids

and eye lashes (Fig. 3).

ICE2005 and ICE2006 Databases

The Iris Challenge Evaluation (ICE) was conducted and

managed by the National Institute of Standards and

Technology (NIST) and consisted of an iris recognition

challenge problem distributed to various participating

universities, government agencies, and biometric tech-

nology companies [7]. The broad goals of ICE were

to facilitate the development of iris recognition tech-

nology along with assessing the state of iris recognition

systems.

The ICE2005 database consists of 2,953 640  480

intensity images with varying numbers of images ac-

quired from each subject [8]. The images are 8-bit gray-

level in the TIFF image format. Exactly 1,425 images

Iris Databases. Figure 2 Example images from UPOL iris

image database.

Iris Databases. Figure 3 Image from the Bath iris-image

database demonstrating the illumination array centered

in the pupil [6].

Iris Databases

771

were captured from the right eye of 124 subjects and

1,528 images were captured from the left eye of 120

subjects. Images of both eyes of 112 subjects were

collected resulting in 132 total subjects used to con-

struct the database. Images were collected using an LG

Iris Access 2200 iris camera. There are representations

of various types of image noise within the database

such as iris occlusion, poor focus, and partially cap-

tured eyes as illu strated in Fig. 4. A larger databas e of

over 65,000 images captured from 356 subjects in

planned for release. The new database is a superset of

the ICE2005 database and the later constructed

ICE2006 database.

MMU1 and MMU2 Databases

The Multimedia University constructed a relatively

small data set of 450 320  240 grayscale iris images

designated as MMU1 [9]. These images were captured

using the LG IrisAccess 2200 camera. Examples images

are shown in Fig. 5. Information regarding the number

of eyes and subjects used to construct this database is

not provided.

A new database of 995 iris images captured from

100 subjects from Asia, Middle East, Africa, and

Europe, using the Panasonic BM-ET100US Authenti-

cam. Five images were taken from each of the subjects’

eyes. Five left eye images were discarded due to a

subject’s diagnosis of cataract disease. As with many

of the previous databases, noise present in the images

was mainly due to iris occlusion.

WVU Non-Ideal Iris and Off Axis/Angle

Databases

The West Virginia University constructed two iris-

image databases [10, 11]. The iris portion of the data-

base consists of 3,099 480  640 grayscale iris images

captured from 244 subjects. The images were captured

during a single session using the OKI IrisPass-H sen-

sor. The number of images from each eye varies

between three and six. This database was constructed

as a non-ideal iris-image database, and therefore, con-

tains a considerable number of images that depict

various types of noise, which could be encountered

in a real world scenario. Noise types include iris occlu-

sions, varying illumination, poor focus, and off-angled

images. An example image from this database is shown

in Fig. 6(a).

Most iris databases are composed of frontal view

iris images. The second WVU iris database is com-

posed of images of irises taken at various angles. The

database was captured using two sensors. The ﬁrst set

of images was captured using the Sony Cyber Shot

DSC F717 camera used in infrared mode. There are

268 2560  1920 RGB images captured from 19 sub-

jects during a single acquisition session. During data

collection the camera was positioned at the angles of

0, 15, and 30 degrees. The second set of images for this

database was captured using a monochrome camera.

Iris Databases. Figure 4 Example images from ICE iris

database.

Iris Databases. Figure 5 Example images from MMU1 iris

database.

Iris Databases. Figure 6 Example images from WVU

non-ideal iris image and off-axis/angle iris image

databases. (a) Non-ideal iris image example (b) Off-angle

iris image example.

772

Iris Databases

There are 597 720  480 intensity images captured

from 73 subjects during a single acquisition session.

As with the other image set, the camera was positioned

at the angles of 0, 15, and 30 degrees during data

collection. An example iris image captured at 15

degrees off-angle is shown in Fig. 6(b).

UBIRIS.v1 and UBIRIS.v2 Database

In order for researchers to test the robustness of

iris segment ation/recognition algorithms when using

noisy images, University of Beira Interior constructed

an iris database of noisy iris images [12, 13]. The

images were captured using the Nikon E5700 camera.

There are 1,877 2560  1704 RGB images captured

from 241 subjects during two acquisition sessions. To

introduce noise to the process, the location of the

acquisition session was changed to facilitate changes

in natural luminosity, contrast, reﬂections, and focus,

as shown in Fig. 7.

Recently, the University of Beira Interior made

available a second version of the UBIRIS database,

UBIRIS.v2, which currently consists of 11,000 images

and is only available to participants in the

▶ Noisy Iris

Challenge Evaluation (NICE) Part I contest [14].

Summary

Iris-image databases are crucial to the development

and advancement of iris-based biometrics . These data-

bases along with prescribed evaluation methodologies

allows for direct comparison of iris segmentation/rec-

ognition algorithm performance. The databases will

increase in size and complexity of iris-image until all

algorithmic problems, inefﬁciencies, and shortcomings

have been fully addressed.

Related Entries

▶ Iris

▶ Iris Acquisition Device

▶ Iris Challenge Evaluation

▶ Iris Image Quality

▶ Iris Sample Synthesis

Bibliography

1. Institute of Automation, Chinese Academy of Science: CASIA

v1.0 Iris Image Database, 2008. http://www.nlpr.ia.ac.cn/english/

irds/irisdatabase.htm. Accessed 27 Dec, 2008

2. Phillips, P.J., Bowyer, K.W., Flynn, P.J.: Comment on the CASIA

version 1.0 Iris Dataset, IEEE Trans. Pattern Anal. Mach. Intell.

29(10), 1869-1870 (2007)

3. Institute of Automation, Chinese Academy of Science: CASIA

v3.0 Iris Image Database, 2008. http://www.nlpr.ia.ac.cn/english/

irds/irisdatabase.htm. Accessed 27 Dec, 2008

4. Dobes

, M., Machala, L.: UPOL Iris Image Database, 2008. http://

phoenix.inf.upol.cz/iris/. Accessed 27 Dec, 2008

5. Dobes

, M., Machala, L., Tichavsky

, P., Pospı

il J.: Human Eye Iris

Recognition Using the Mutual Information. Optik 115(9),

399–405 (2004)

6. University of Bath: University of Bath Iris Image Database, 2008.

http://www.bath.ac.uk/eleceng/research/sipg/irisweb/index.html.

Accessed 27 Dec, 2008

7. National Institute of Standards and Technology: Iris Challenge

Evaluation (ICE), 2008. http://iris.nist.gov/ICE/. Accessed 27

Dec, 2008

8. Liu, X., Bowyer, K.W., Flynn, P.J.: Iris Recognition and Veriﬁca-

tion Experiments with Improved Segmentation Method. In Pro-

ceedings of Fourth IEEE Workshop on Automatic Identiﬁcation

Advanced Technologies (AutoID). Buffalo, NY, 17–18 October

2005

9. Multimedia University: MMU1 and MMU2 Iris Image Data-

bases, 2008. http://pesona.mmu.edu.my/~ccteo. Accessed 27

Dec, 2008

10. West Virginia University: West Virginia University Biometric

Dataset Collections, 2008. http://www.csee.wvu.edu/~simonac/

CITeR_DB. Accessed 27 Dec, 2008

11. Ross, A., Crihalmeanu, S., Hornak, L., Schuckers, S.: A Centra-

lized Web-Enabled Multimodal Biometric Database. In Proceed-

ings of the 2004 Biometric Consortium Conference (BCC),

Arlington, VA, September 2004

12. Proenc¸a, H., Alexandre, L.: UBIRIS: A Noisy Iris Image

Database. In: Proceedings of the 13th International Conference

on Image Analysis and Processing (ICIA2005), Vol. 1, pp. 970–

977, 2005

13. SOCIA Lab – University of Beira Interior: UBIRIS.v1 Iris Image

Database, 2008. http://iris.di.ubi.pt/ubiris1.html. Accessed 27

Dec, 2008

14. SOCIA Lab – University of Beira Interior: Noisy Iris Challenge

Evaluation – Part I, 2008. http://nice1.di.ubi.pt/

Iris Databases. Figure 7 Example images from UBIRIS.v1

iris database [12].

Iris Databases

773

Iris Device

JAMES R. MATEY

Electrical and Computer Engineering Department,

US Naval Academy, Annapolis, MD, USA

Synonyms

Iris camera; Iris image capture device; Iris reader; Iris

camera; Iris scan ner

Definition

An iris device is a device that acquires images of the iris

for use in biometric recognition. More commonly re-

ferred to as iris cameras, readers or scanners, these

devices typically include some form of active

▶ near

infrared illumination (NIR), since the current iris rec-

ognition algorithms were designed for NIR images of

the eye. In addition, the device may include features to

aid subjects in properly aligning their eyes in the ﬁeld

of view, spooﬁng countermeasures, and on board pro-

cessing of the acquired images.

Introduction

Iris recognition is one of the strongest biometrics

available [1–4]. Iris recognition is a strong biometric

because: (1) the human iris is a complex structure with

a high degree of randomness; (2) the iris is protected;

(3) the iris is accessible; and (4) the structures of the

iris that are used for iris recognition are stable, from

early childhood on – in the absence of illness or injury

that disrupts the iris tissue.

The ﬁrst mention of iris patterns as a biometric w as

likely a paper by Bertillon [5]; several others subse-

quently suggested iris patterns as a biometric and the

idea was a plot element in the 1983 James Bond ﬁlm

Never Say Never Again [6]. However, it was not until

the early 1990’s that John Daugman developed a prac-

tical algorithm for iris recognition based on Gabor

wavelets [7]. Minor variants on the Daugman algo-

rithm remain the dominant algorithms in commercial

iris recognition systems as of 2008, though there are

vigorous research efforts into alternative algorithms

[8]. The commonly used name for the Daugman algo-

rithms in current use is

▶ iris2pi.

Further discussion of algorithms may be found in

other entries that cover the algorithmic aspects of iris

recognition in detail. This entry is concerned with

devices that can acquire images suitable for iris recog-

nition. The reader may well ask, ‘‘What is a suit able

image?’’ That question depends on the algorithms used

for recognition, and the determination of the mini-

mum quality image for a given application is still a

matter for research. However,

▶ ANSI and ▶ ISO have

published standards [9] for iris image quality that are

generally accepted for commercial applications. From

the standpoint of iris image capture device design, the

most impor tant of these image quality metrics are



▶

Resolution: 100–200 ▶ pixels across the iris



Signal to noise ratio: 40 dB, 100:1. For 8 bit pixels

exercised over the full dynamic range, the noise

would be 2 digita l numbers (DN). Equiva-

lently, 7 bits should be meaningful.



▶

Contrast: For 8 bit pixels, 90 gray levels of sepa-

ration between the iris and sclera and a 50 gray

levels of separation between iris and pupil



Operating ▶ wavelength: 700–900 nm

Brief History of Iris Image Acquisition

Devices

The ﬁrst commercial iris image acquisition device was

the System 2000 from IrisScan [ 10] in 1995. In the past

decade numerous iris image acquisition devices have

been introduced, as can be seen from the partial listing

in Table 1. To the author’s knowledge, there has not yet

been a comprehensive, publicly available evaluation of

the relative merits of these devices. One of the most

comprehensive tests to date was conducted by the

International Biometrics Group (IBG) for the US

government. There have also been tests of iris recogni-

tion algorithms – rather than the acquisition devices –

such as the Iris Challenge Evaluation (ICE) competi-

tion sponsored by the NIST.

The IBG study compared one product each

from Panasonic, OKI and LG. A subsequent IBG

study [11] reported on the IrisGuard H100. The lack

of more comprehensive iris device evaluatio ns is

understandable. Iris recognition works quite well;

its failure rates are low. Hence, statistically signiﬁcant

774

Iris Device

probes of the false match, false non-match, failure to

enroll and failure to acquire rates require many indi-

vidual trials, making such tests difﬁcult and expensive.

The addition of evaluations of ease of use and suitabil-

ity for various scenarios only compounds the problem.

Another non-technical issue is also important: the

iris recognition marketplace is intensely competitive –

so much that some of the rivals have had protracted

legal battles in the courts. Under these conditions,

convincing a vendor to participate in a public test

whose results might show that their product is in

some way inferior to a competitor’s product can

be far more difﬁcult than any technical challenge.

There have been unpublished evaluations of iris

devices by government and private industry – some

of which may be obtained by request to the device

manufacturers.

This entry will trace the operation of a generic

iris image acquisition device – from

▶ photons to

identity – and di scuss the metrics that are important

for evaluation of such devices in generic dep loyment

scenarios.

Photons to Iris Image

Iris recognition starts with photons of light from am-

bient illumination or from an active illuminator within

the iris recognition system. The photons impinge upon

the subject iris and are partially reﬂected. The reﬂected

photons fall into two categories, specular

▶ reﬂection

and diffuse reﬂection. The specular reﬂection is due to

the impedance mismatch between eye tissue and air;

the physics is the same as that for the reﬂections that

you see in a piece of otherwise clear window glass or a

bathroom mirror. Since the eye is curved, it acts as a

convex mirror and reduces the apparent size of objects

seen in its specular reﬂections (Some automobile rear

Iris Device. Table 1 Partial List of Iris Image Acquisition Devices

Vendor Model Year introduced

IrisScan System 2000 1995

OKI IrisPass

-S 1998

LG 2200 1999

Sensar R1 1999

Iridian Authenticam ™ 2000

Panasonic BM-ET-100 – Authenticam™ 2001

LG 3000 2002

OKI IrisPass

-WG 2002

Panasonic BM-ET-500 2002

IrisGuard H-100 2003

Securimetrics Pier™ 2.2 2003

Securimetrics Pier™ 2.3 2003

OKI IrisPass-H 2004

Panasonic BM-ET-300 2004

LG 4000 2005

OKI IrisPass

-M 2005

Sarnoff Corporation Iris On the Move™ 2005

Securimetrics HIIDE™ 2005

IriTech Neoris 2000 2006

Jiris JCP1000 2006

Panasonic BM-ET-330 2006

Hoyos Hbox™ 2007

Panasonic BM-ET-200 2007

IrisGuard AD-100 2008

Iris Device

775

view mirrors are convex and carr y the warning

‘‘Objects may be closer than they seem’’).

The diffusely reﬂected fraction is the

▶ albedo of

the iris and is of the order of 10% in the near infrared

(NIR). The diffusely reﬂected fraction varies across the

iris – that is what gives rise to the patterns seen in an

iris image. The diffusely reﬂected photons are scattered

in all directions. The lens of the iris camera will inter-

cept a small fraction of the scattered photons; the

fraction is, to ﬁrst order, the area of the lens divided

by the area of a half-sphere centered on the subject

with a radius equal to the camera-to-subject

▶ standoff

distance. Mathematically, the fraction is

lens

camerasubject



;

where r is the lens radius and x is the camera to subject

distance. A lens 1 cm in diameter at a distance of 10 cm

will capture approximately 0.1% of the light diffusely

scattered from an iris (or any other object).

The lens will focus an image of the iris, and the

variation of its diffuse reﬂection, onto an imaging

sensor. The sensor could be a piece of photographic

ﬁlm. In practice, for iris imaging devices it is almost

always a piece of silicon in the form of either a

▶ CCD

imager or a

▶ CMOS imager. The most important

characteristics of the imager are its pixel size, the num-

ber of pixels, and the

▶ quantum efﬁciency (QE) of

the pixels.

The pixel size of the imager sets the

▶ focal length

of the lens. The iris is approximately 1 cm across for

most of the population. ANSI standards require 100 to

200 pixels across the iris; this corresponds to a pixel

size at the iris of the order of 100 microns. The pixel

size at the imager is of the order of 10 microns. Hence,

the lens system of the camera must magni fy (minify)

the iris by a factor of approximately 0.1. A desired

resolution at the iris and a known imager pixel size

deﬁne the

▶ magniﬁcation required. The magniﬁca-

tion and the camera-to-subject standoff distance then

specify the focal length of the camera lens through the

simple lens equation

M ¼

;

where M is the magniﬁcation, f is the focal length, and

p and q are the subject-to-lens and lens-to-se nsor dis-

tances respectively. For a magniﬁcation of 0.1 and a

standoff distance of 2 meters, the lens will have a focal

length of ~200 mm. Note that the smallest possible

lens-to-sensor distance, q, is f. Hence these considera-

tions also drive the minimum size of the camera/lens

package.

The number of pixels in the imager sets the ﬁeld of

view of the imager. For 100 pixels across the iris, the

minimum is 100 pixels across the imager – assuming

perfect alignment of the subject iris with the camera. In

practice, rather more is required. Camera s designed for

200 pixels across iris frequently use imagers that are

640  480 – well more than twice the minimum

number of pixels. Cameras designed for large stand-

offs, such as the Iris on the Move™ system, use much

larger (2048  2048) imagers to relieve the alignment

requirements on both the subject and the system.

The quantum efﬁciency of the sensor is the fraction

of photons that are converted into electrons. More

electrons give a larger signal and bette r

▶ signal to

noise ratio (SNR). The number of electrons is just the

quantum efﬁciency times the number of photons de-

livered to the sensor pixel. The primary noise sources

for sensors used in iris systems are

▶ read noise and

▶ shot noise. Read noise is introduced each time a

pixel is read; the read noise power is to ﬁrst order

constant for a given camera conﬁguration. Shot noise

is the result of electron/photon statistics – the random-

ness with which photons arrive and are converted to

electrons. Shot noise is propor tional to the square root

of the number of electrons. Iris sensors normally oper-

ate in a shot noise dominated regime, so doubling the

quantum efﬁciency of the sensor will improve the SNR

by the square root of 2, rather than 2. The number of

photons depends on both the rate of delivery and the

shutter time of the camera. Matey et al publi shed an

analysis of the tradeoffs between SNR and other sys-

tems design decisions [12].

If there were no concern about the safety of the

subject, the SNR of the system could be made arbi-

trarily large by simply increasing the illumination level

on the subject iris. However, the eye is sensitive to

illumination – even illumination that it cannot see.

The American Conference of Governmental Industrial

Hygienists (ACGIH) provides guidelines,

▶ threshold

limit values (TLVs), for acceptable exposures to

776

Iris Device

infrared illumination [13]. The guidelines cover both

the irradiance (W/cm

) at the eye as well as the radi-

ance (W/sr-cm

) of the source. Any iris imaging system

needs to take these guidelines as well as the various

national and international safety standards and regula-

tions into account.

In summary, photons travel from the illuminator

to the eye and are reﬂected, specularly and diffusely, to

the camera lens. The camera lens focuses the photons

onto the camera sensor; the sensor converts the

photons to electrons; the number of electrons is

measured and the meas urement is converted to a digi-

tal signal that is then assembled into a digital image for

subsequent processing by a biometric recognition al-

gorithm that can determine if the eye in front of the

camera has previously been enrolled in the database

against which current eye is being compared.

Iris Image Acquisition Device System

Metrics

Resolution, SNR, contrast and operating wavelength

and the resulting image quality are crucial metrics for

the iris images, as noted in the introduction. However,

iris image acquisition devices must ﬁt into systems

deployed in the real world – and in the real world,

those metrics are not enough.

Ease of use, robustness, reliabilit y,

▶ interoperabil-

ity and cost are several of the real world systems

metrics that are important considerations for iris ac-

quisition devices. Each of these could be the topic of an

entry on its own. The article now brieﬂy discusses ease

of use.

In the author’s opinion, the primary factors for

ease of use are

▶ capture volume, ▶ residence time

and subject motion. Capture volume is the volume

over which a good quality iris image can be reliably

captured. Small volumes make it difﬁcult for subjects to

present their irises to the system. Residence time is

the length of time that a subject must hold their iris

within the capture volume. Small volumes with long

residence times are particu larly difﬁcult. Large volumes

with short residence times are almost always better.

Subject motion is complicated. How much motion

is tolerable depends on direction, longitudinal or trans-

verse to the camera line of sight. Tolerance to motion

can be improved by using short shutter times, at the

expense of SNR. In general, systems that allow for

subject motion are easier for the subjects to use.

Ease of use can be traded off in some circum-

stances. If the users of a system are well trained and

habituated, the system can be successful, even if it is

very difﬁcult for a ﬁrst time user.

Related Entries

▶ Biometric Data Interchange Format, Standardization

▶ Biometric System Design

▶ Iris Encoding and Recognition

▶ Iris Image Qu ality

▶ Iris Recognition, Overview

References

1. Mansﬁeld, A., Kelly, G., Chandler, D. Kane, J.: ‘‘Biometric Prod-

uct Testing: Final Report’’ (CESG Contract X92A/4009309),

Centre for Mathematics and Scientiﬁc Computing, UK National

Physical Laboratory (2001)

2. Daugman, J.: ‘‘How Iris Recognition Works’’, IEEE Trans. Circ.

Syst Video Tech. 14(1), (2004)

3. International Biometrics Group, ‘‘Independent Testing of Iris

Recognition Technology, Final Report, May 2005’’,

NBCHC030114/0002. Study commissioned by the US Depart-

ment of Homel and Security

4. Jain, A.K.: Iris recognition. In: Jain, A.K., Flynn, P.J., Ross, A.

(eds.) Handbook of Biometrics, Springer, New York (2007)

5. Bertillon, A.: ‘‘La couleur de L’Iris’’ Annales de Demographie

Internationale 7 226–246 (1886)

6. Never Say Never Again [motion picture], Schwartzman, J. pro-

ducer. Warner Bros. 1983

7. Daugman, J.: Biometric Personal Identiﬁcation System Based on

Iris Analysis. U.S. Patent No. 5,291,560 (1 March 1994)

8. Phillips, P., Jonathon, W., Scruggs, T., Alice J.: O’Toole, Flynn, P. J.,

Bowyer, K. W., Schott, C. L., Sharpe, M.: ‘‘FRVT 2006 and ICE

2006 Large-Scale Results.’’ NISTIR 7408, March 2007. See also

http://iris.nist.gov/

9. ANSI INCITS 379-2004. Iris Image Interchange Format,

American National Standards Institute, Inc., New York, NY

10. IrisScan and Sensar were merged to form Iridian. Iridian was

subsequently bought by Viisage (now L1 Identity Solutions) and

merged into their Securimetrics division

11. International Biometrics Group, ‘‘Comparative Biometric Test-

ing, Round 6 Public Report’’, September 2006

12. Matey, J.R., Ackerman, D., Bergen, J., Tinker, M.: Iris recog-

nition in less constrained environments. In: Ratha, N. and

Govindaraju, V. (eds.) Advances in Biometrics, Sensors,

Algorithms & Systems, Springer, London (2008)

13. ACGIH, 2006 TLVs and BEIs, ISBN 1-882417-62-3, www.

acgih.org

Iris Device

777

Iris Digital Watermarking

NICK BARTLOW,NATHAN KALKA,BOJAN CUKIC,

RUN ROSS

West Virginia University, Morgantown, WV 26506,

USA

Synonyms

Biometric Watermarking; Digital Watermarking

Definition

Iris digital watermarking is a speciﬁc type of ▶ digital

watermarking that involves the imperceptible embed-

ding of data (the watermark) in iris images (the host)

in order to imp art additional security to an iris bio-

metric system. The watermark data can be strings of

random digits; system speciﬁc identiﬁers, such as or-

ganization names and ﬁle creation dates; or biometric

feature vectors. The additional security offered by the

watermark may result from using it as a mechanism for

proving the authenticity of the host image, trackin g the

chain of custody, or incorporating a multimodal bio-

metric option. Iris digital watermarking or

▶ biomet-

ric watermarking, in general, is typically used in

tandem with

▶ cryptography, and import antly, pro-

vides a layer of security that remains intact after the

decr yption process. A functional iris digital watermark

should not degrade the performance of the host bio-

metric system it protects.

Introduction

Coined by Tirkel et al. [1], the term ‘‘digital water-

mark’’ originated in 1993. Unlike their physical pre-

decessors (i.e., currency, copyright marks, etc.), digital

watermarks are usually imperceptible to the human

eye, requiring the use of machines for detection and

extraction from the host media in which they are

embedded. Digital watermarking is closely related to

the ﬁeld of

▶ steganography where secret messages are

clandestinely embedded in larger, unrelated messages

[2]. The beneﬁts of digital watermarking are somewhat

broad and can fall under one or more areas of interest.

Traditionally, digital watermarking has been employed

as a form of copyright protection that allows an

individual to prove (or disprove) ownership by embed-

ding and extracting data suitable for veriﬁcation. Ad-

ditionally, digital watermarking can be applied to

verify the authenticity of digital media, provide copy-

right protection or reproduction management, and

offer another mechanism for content description [3,

4]. Biometric watermarking is a particular case of

digital watermarking where the content of the water-

mark or the host data (or both) are biometric entities.

This imparts an additional layer of authentication to

the underlying system. Figure 1 shows examples of the

four main classes of digital watermarking.

A digital watermark can be characterized using attri-

butes such as visibility, blindess, and symmetry. Kutter

provides a comprehensive description of these attri-

butes athttp://www.watermarkingworld.org/. Forma-

lized deﬁnitions of these attributes are outlined in

Table 1. The issue of visibility or perceptibility relates

to whether or not the watermark is noticeable by

humans. The issue of blindness indicates whether or

not the process of detection and extraction of the

watermark relies on the original host data or other

auxiliary data. Auxiliary data in semiblind systems

can include information dealing with encoding mod-

ule input parameters, encoding locatio ns, or any other

information used to assist in the detection and or

extraction processes. Finally, watermarking systems

must make use of keys, either private similar to sym-

metric cryptographic systems or public key pairs akin

to asymmetric cryptosystems. Although biometric

watermarking system characterized by any combina-

tion of these attributes is plausible, an invisible, blind,

and asymmetric system is arguably the most difﬁcult to

conceive.

The quality of a biometric watermarking system

can be evaluated based on ﬁve measures: impercept-

ibility, robustness, fragility, capacity, and performabil-

ity. Table 2 describes each of these ﬁve measures in

depth. Although the characteristics imperceptibility,

robustness, fragility, and capacity apply to digital

watermarking, performability is speciﬁc to biometric

watermarking.

Algorithms

Overview

A generalized iris digital watermarking framework can

be broken into three main operating modules:

778

Iris Digital Watermarking

Iris Digital Watermarking. Table 1 Description of digital watermarking types applicable to biometric watermarking

Type Description

Visible The embedded information is noticeable by humans, either visually in pictures or

audibly in sound files

Invisible The embedded information is either completely imperceptible or not easily

noticeable by humans without the assistance of machines

Public (blind) The original host file is not required to detect/extract the embedded watermark

Semiblind The embedded watermark is detected with additional information

relative to the watermark encoding scheme, but does not require the

entire original host file

Private (nonblind) The embedded watermark can only be detected/extracted with both the watermarked

image and the original host file

Symmetric (Private Key) A secret / private key is utilized to encode the watermark in the

host image. This requires communication of the secret key between the

sender and the receiver

Asymmetric (Public Key) A public–private key pair is used to encode the watermark in the host file.

The use of a public key pair prevents the need to communicate a secret key

between sender and the receiver, as only the public portions of the key pairs

need to be available. Optionally, this method can ensure image integrity and

nonrepudiation of origin

Iris Digital Watermarking. Figure 1 The four classes of digital watermarking (a) Biometric watermark embedded in

biometric host data; (b) Biometric watermark embedded in nonbiometric host data; (c) Nonbiometric watermark

embedded in biometric host data; (d) Nonbiometric watermark embedded in nonbiometric host data. Cases (a)–(c) are

examples of biometric watermarking. The images of Lena and the Mandrill are reproduced from the USC-SIPI Image

Database.

Iris Digital Watermarking

779