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

Подождите немного. Документ загружается.

Related Entries

▶ Biometric Data Interchange Format, Standardization

▶ Biometric Sensor and Device, Overview

▶ Biometric System Design

▶ Iris Acquisition Device

▶ Iris Encoding and Recognition

▶ Iris Image Quality

▶ Iris Recognition, Overview

▶ Iris Super-Resolution

References

1. Matey, J.R., Naroditsky, O., Hanna, K., Kolczynski, R., LoIacono,

D., Mangru, S., Tinker, M., Zappia, T., Zhao, W.Y.: Iris on the

Move™: acquisition of images for iris recognition in less con-

strained environments. Proc. IEEE 94(11), 1936–1947 (2006).

2. Mansﬁeld, A., Kelly, G., Chandler, D., Kane, J.: ‘‘Biometric prod-

uct testing: ﬁnal report’’ (CESG Contract X92A/4009309), Cen-

tre for Mathematics and Scientiﬁc Computing, UK National

Physical Laboratory (2001)

3. Daugman, J.: How iris recognition works. IEEE Trans. Circuits

Syst. Video Technol. 14(1), 21–30 (2004)

4. International Biometrics Group: Independent testing of iris rec-

ognition technology, Final report, May 2005, NBCHC030114/

0002. Study commissioned by the US Department of Homeland

Security

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

Internationale 7 , 226–246 (1886)

6. Schwartzman, J. producer.: Never Say Never Again [motion

picture]. Warner Bros. (1983)

7. Daugman, J.: ‘‘Biometric personal identiﬁcation system based on

iris analysis.’’ US Patent 5,291,560, 1 Mar 1994

8. Phillips, P.J., Scruggs, W.T., O’Toole, A.J., 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. Daugman, J.: Probing the uniqueness and randomness of Iris-

Codes: results from 200 billion iris pair comparisons. Proc. IEEE

94(11), 1927–1935 (2006)

10. ISO/IEC: ‘‘Biometric data interchange formats – Part 6: Iris

image data’’ ISO/IEC JTC 1/SC 37 N 504 (2004)

11. Smith, W.J.: Modern Optical Engineering . McGraw-Hill, New

York (2000). ISBN 0-07-136360-2. http://en.wikipedia.org/wiki/

Depth_of_ﬁeld

12. ACGIH: TLVs and BEIs (2006). ISBN 1-882417-62-3. www.

acgih.org

Iris Quality Metrics

▶ Iris Image Quality

Iris Reader

▶ Iris Acquisition Device

▶ Iris Device

Iris Recognition, Overview

YUNG-HUI LI

,MARIOS SAV VID ES

Language Technolog y Institute, Carnegie Mellon

University, Pittsbu rgh, PA, USA

Department of Electrical and Computer Engineering,

Carnegie Mellon University, Pittsburgh, PA, USA

Synonym

Iris Biometric

Definition

Iris recognition emerges as one of the most useful

modalities for biometrics recognition in last few dec-

ades. The goal of iris recognition is to recognize human

identity through the textural characteristics of one’s

iris muscular patterns. The procedures for iris recogni-

tion usually consist of four stages: image acquisition,

iris segmentation, feature extraction, and pattern

matching. The iris recognition has been acknowledged

as one of the most accurate biometric modalities

because of its high recognition rate. It has been applied

in the ﬁeld of border control and national security.

More and more countries and private companies

have shown interests to use the technique of iris recog-

nition. Large scale application of iris recognition in

daily life is just a matter of time.

Introduction

The goal of biometric recognition is to recognize

human identity by comparing the features of their

physiological or behavioral characteristics. There are

dozens of such characteristics that can be found and

used for such purpose. Among them, ﬁngerprint, face,

and voice are most famili ar to most people. However,

810

Iris Quality Metrics

the performance (in terms of the recognition rates) for

all the above modalities is not good enough for the

purpose of applications that require high level of

security.

Not very long ago, identity recognition based on

iris texture patterns is proposed as another modality

for biometric recognition [1–3]. Iris is the area that

appears as an annular shape between the pupil and the

sclera, as illustrated in Fig. 1. There are many advan-

tages of using iris as the modality of biometric recog-

nition. For example, iris pattern is developed and

becoming stable after the eighth month of gestation.

Besides, iris is well-protected by cornea, not easily

changed by external factors. Therefore, iris patterns

are very stable, relative to other modalities such as

face or voice, and very suitable to be used as a unique

feature for identity recognition. Another advan tage is

that randomness of the muscle of the iris region makes

possibility of sharing exactly the same iris pattern

among different persons very low, which provides

high discriminability for this feature.

A typical iris recognition system consists of the

following stages: iris image acquisition, iris segmen-

tation, feature extraction, and pattern matching, as

illustrated in Fig. 2. Image acquisition is of vital

importance in iris recognition performance because

the better the image quality, the more accurate result

the system can achieve. After iris image is acquired, the

next step is to locate the position of the iris and pupil.

Several techniques have been proposed and most of

them are based on image gradient and edge detection.

After iris has be en located, features of the iris textures

can be extracted. What feature to extract and how to

extract them is also important issue in iris recognition.

Good features give high inter-class variation and low

within-class variation, which is desired in common

biometric recognition system. Finally, one should be

able to compare features from different irises and ob-

tain a score of similarity which tells how similar these

two irises are (or a score of distance, which tells how

much different they are). Decision of classifying test

image as authentic or imposter is made by comparing

the similarity score with a given threshold.

Iris Image Acquisition

The area of iris is very small compared with the area of

the whole face. Therefore, it is not a trivial task to take

a clear iris image with high quality. On the other hand,

user-friendliness is also an important factor which

has to be considered in the process of iris image

acquisition. According to the user-friendliness and

the image quality, iris image acquisition device can be

categorized into three different groups. They are tradi-

tional iris acquisition device, middle distance iris ac-

quisition device, and long distance iris acquisition

device. Each of them is introduced in the following

paragraphs.

For traditional iris acquisition devices, in order to

take iris image s that contain enough useful detailed

Iris Recognition, Overview. Figure 2 Flow chart of a

typical iris recognition system.

Iris Recognition, Overview. Figure 1 Illustration of the

areas of pupil, iris and sclera.

Iris Recognition, Overview

811

information, people usually use cameras which

has specialized lens set up on a ﬁxed stand. At the

same time, there is also a ﬁxed stand for subjects to

lean their heads against. In this way, the system mini-

mized the possibilit y of subtle movement of subject’s

head during the iris image acquisition stage, and max-

imized the quality of the acquired iris image. Exemplar

images of traditional iris acquisition device are

shown in Fig. 3.

Traditional iris acquisition devices are good for

capturing high quality iris image, but extremely inﬂex-

ible and not user-friendly. It requires high level of user-

cooperation to successfully acquire iris image. Is it

possible to lower the level of required user-cooperation

and at the same time, maintain the high quality of the

eye image? Recently, progress of innovation in iris

acquisition device has made it possible to lessen the

requirement of user-cooperation while maintaining

high quality iris imag es. They can be categorized as

middle distance iris acquisition devices because they

can capture users’ iris images even when users stand at

a distance of 50–100 cm away from the camera. Users

are not required to put their heads against a rack.

Therefore, it is more user-friendly and convenient for

general public.

One of examples of middle distance iris acquisition

devices is the LG iCAM4000. The functional unit of LG

iCAM4000 is shown in Fig. 4, and an example picture

is shown when it is used to acquire a user’s iris image.

Long distance iris acquisition devices ar e able to

captureirisimagesevenwhenusersarestandingata

further distance, or even when they are moving. One

example of this kind of devices is the iris-on-the-move™

(IOM) manufactured by Sarnoff Corporation. The sys-

tem provides a portal which subjects are expected to

walk through. At the end of the walking aisle, four

high resolution infrared (IR) camera s are placed inside

a cabinet. During the time subjects are walking in the

portal, IR illuminators located at the wall of the portal

radiate IR light, at the same time, high resolution

cameras which tuned to sync with the illuminators

take continuous shots of the subjects. Then a face and

eye detector is applied on the image to ﬁnd the loca-

tion of the eyes. The whole iris acquisition process is

fully automated and it only requires subjects to walk

through the portal, w ithout standing still or placing

Iris Recognition, Overview. Figure 3 (a) Example of traditional iris acquisition device (b) Example picture to illustrate

how to use it.

Iris Recognition, Overview. Figure 4 (a) LG iCAM4000 iris camera (b) Example picture to show how to use it.

812

Iris Recognition, Overview

their heads at a ﬁxed position. In this way, the system

alleviates the inconvenience of the tr aditional iris

acquisition devices while being able to maintain a

reasonable quality of the eye pictures.

Figure 5 shows pictures of IOM system.

It is very intuitive that the system which requires

least user cooperation (more user-friendly) has the

highest probability of getting eye images of lowest

quality. For example, the IOM system is very user-

friendly. It does not require users to stand still at a

particular point to get their iris images. However, the

eye image s taken from IOM are very easy to be blurred

or out of focus. This is because the images may be

taken when the subjects are walking in a region which

is outside of the depth of the ﬁeld of the cameras.

Recently, a new camera system design has been

proposed to address this issue. Dowski and Johnson

proposed a low cost imaging system which combines

the nonrotationally symmetric aspheric optical ele-

ments and digital signal processing in a fundamen-

tal manner to vastly extend the depth of ﬁeld of

the imaging systems [4]. It is called ‘‘Wavefront Cod-

ing’’ imaging system. The block diagram of the Wave-

front Coding system is shown in Fig. 6. The optical

section is modiﬁed by adding a generalized aspheric

Wavefront Coding optical element near the aperture

stop. This will make the images formed on CCD to

have a specially well-deﬁned blur that is insensitive to

misfocus. Digital image processing technique is ap-

plied to the blurred image to reproduce the sharp

and clear image. Optical systems aided with Wavefront

Coding are able to reproduce a clear and focused image

even when the subject is standing at the out of focus

zone. For more details, please refer to [4].

Iris Segmentation

No matter which kind of iris acquisition device is

used, the eye image taken usually looks like the one

in Fig. 1. Therefore, it is necessary to locate the iris

region in this imag e and focus our future process

Iris Recognition, Overview. Figure 5 (a) IOM system. Camera cabinet is on the left; the portal is on the right, behind the

subject. (b) Example picture of subject walking through the portal.

Iris Recognition, Overview. Figure 6 Block diagram of Wavefront Coding imaging system.

Iris Recognition, Overview

813

(feature extraction and pattern matching) on this re-

gion alone. The process of locating iris region is called

iris segmentation.

As stated above, iris region is of annular shape.

Therefore, it is intuitive to segment iris with two

circles: one circle indicates the boundary between

iris and pupil, and the other indicates the boundary

between iris and sclera. The texture in the region

between these two circles is exactly what we need for

further process.

There exist different algorithms which ser ve the

same goal of iris segmentation. One of the most popu-

lar algorithms is proposed by Wildes [5]. There are two

steps in this algorithm. The ﬁrst step is to apply a

gradient-based edge detector on the whole eye image

to generate an edge-map. The edge-map tells us the

position where strong edges exist (strong differences in

pixel values). Intuitively, those positions are possible

candidates of the iris boundaries since the two bound-

aries of iris, both inner and outer, are the positions

where high pixel contrast takes place. The second step

is to ﬁnd the exact two boundaries from this edge-

map. The method Wildes used is a voting scheme.

Every circle on a 2D plane can be described by three

parameters, coordinate in x and y axis, and the radius r.

Therefore, one can construct a three-dimensional

(3D) space, where each dimension represents one

parameter. Every positive pixel on the edge-map can

vote to the point ( x, y, r) in the 3D parameter space as

long as this positive pixel is on the perimeter of the

circle which is parameterized as ( x, y, r). Since different

circles can pass through the same point, it is possible

for 1 pixel on the edge-ma p to vote to multiple points

in 3D parameter space. At last, the point which has the

highest accumulated votes represents the most likely

circle in the original eye image. An example of iris

segmentation is shown in Fig. 7a.

Daugman proposed another segmentation scheme

[6, 7] which is different from what Wildes proposed.

He suggested computing the argument which maxi-

mizes the result of an integrodifferential operator on

the original image. It is described in the following

equation:

max

ðr;x



r;x

Iðx; yÞ

2pr



: ð1Þ

The triple (r, x

) which satisﬁed Eq. 1 is the best

candidate parameter for a circle in the original image.

By applying Eq. 1 twice, we will be able to locate the

boundaries for both iris and pupil. To locate the eyelid

boundary, one only need s to change the form of inte-

grodifferential operator in Eq. 1 from circular to

Iris Recognition, Overview. Figure 7 (a) A raw eye image with iris boundaries identiﬁed as red circles (b) segmented iris

image after masking out unuseful region (c) illustration of the meaning of parameter R and y (d) unwrapped iris image in

polar coordinate.

814

Iris Recognition, Overview

arcuate, with spline parameters, because the shape of

eyelid is most likely to be arcuate.

After observing large number of iris images, one

sooner or later discovers that in fact, the boundaries

of iris and pupil are not exactly circular. They appear as

circular images of low resolution. However, for images

of higher resolution, it is easy to see that using a circular

boundary to describe them is not very accurate.

Therefore, Daugman, in his later work [8], pro-

posed a new segmentation method. He proposed to

use a more ﬂexible model to represent both bound-

aries. This method is called ‘‘active contour,’’ or

‘‘snakes.’’ By using snakes, the boundary of the pupil

and iris is not bounded to circular. They can be of any

shape. Therefore, the discovered boundaries can ﬁt to

the real data more closely, and bring performance

enhancement in the pattern matching stage. The points

of snakes are initialized at the points from sampling N

regularly spaced angular samples of radial gradient

edge data. We have to ﬁrst estimate the Fourier series

expansion of those samples of radial gradient edge

data, as described in the following equation:

N1

y¼0

2piky=N

; ð2Þ

where {r

} is the radial gradient edge data, y =0N–1.

We compute C

from k =0tok = M1. Then we

compute an approximation of the iris boundary {R

}

for y =0(N–1), as described below:

M1

y¼0

2piky=N

: ð3Þ

The number of M controls the smoothness of the

curve. The larger the M, the more ﬂexible the curve is;

and the smaller the M, the less ﬂexible the curve. By

choosing M carefully one can make the discovered

boundaries ﬁt the data more closely and not being

affected by the noises (eyelid or eyelashes).

Most of the time, after iris localization, we would

like to do an image coordinate transformation on the

iris image. The goal of coordinate transformation is to

transform the annular-shaped iris region into rectan-

gular shape, by mapping pixel values from Cartesian

coordinate to polar coordinate. The process is illu-

strated in Fig. 7.InFig. 7b, the region which is unim-

portant for iris recognition is masked out, leaving

only iris texture visible. This iris image is displayed in

Cartesian coordinate. If we pick the center of the pupil

as the origin, and the horizontal line as x-axis, every

point on iris region can be indexed by another coordi-

nate, which is polar coordinate. Polar coordinate is

parameterized with two parameters: R and y, and

the computation of R and y is illus trated in Fig. 7c.If

we take out every pixel in iris region in Fig. 7c and

replot them according to the coordinate ( R, y), we will

get Fig. 7d, which is the iris texture map in polar

coordinate.

There are two advantages of coordinate transfor-

mation. First, the size of the pupil area is not always the

same. It contracts when the ambient lighting is strong

and dilates when the ambient lighting is weak. There-

fore, if we would like to take the iris image in Cartesian

coordinate to perform pattern matching, one problem

is the nonlinear deforma tion of the pattern, which

require much more complex technique to solve. In-

stead, if we perform coordinate transformation for

every iris texture, no matter how much pupil contracts

or dilates, the resulting iris texture map in polar coor-

dinate w ill remain the same. This saves us a lot of

trouble during the later matching stage.

The second problem is, when we take pictures of

eyes, if we are not using traditional iris acquisition

device, it is highly possible that the position of our

camera has relative rotational shift to the subject’s

head. It is very likely to happen because users can freely

move their head in any direction, as long as the eye

images can be captured. If there is indeed relative rota-

tional shift between cameras and the eyes, this is going

to cause another trouble during later matching stage.

This is because matching two patterns with potential

rotational shift is a difﬁcult problem in pattern recog-

nition ﬁeld. Instead, if the iris image is unwrapped

to polar coordinate before featu re extraction and

matching stage, rotational shift in Cartesian coordi-

nate becomes translational shift in horizontal direc-

tion (y axis). Therefore, all we need to do is simply

shift the pattern horizontally in different step sizes

and perform matching again to get correct matching

result.

Feature Extraction

After iris region has been identiﬁed and cropped out,

the next step is to perform feature extraction on the

iris texture. The goal of feature extraction is to extract

the discriminative charac teristic of the iris texture and

store it in a more compact way so that it is more

effective to perform pattern matching in a later stage.

Iris Recognition, Overview

815

Most of the time, in the ﬁeld of image processing

and pattern recognition, the feature extraction is ac-

complished by applying ﬁlters on the input images.

In literature, the ﬁlters which have been used exten-

sively in iris recognition, are Gabor ﬁlters, proposed

by J. Daugman [2, 6, 9, 10, 7]. Gabor ﬁlters can be seen

as complex sinusoids modulated by Gaussian envelope.

They can be expressed as the following equation:

Cðx; yÞ¼A  e





joðcos yÞxjoðsin yÞy

; ð4Þ

where, by convention, the standard deviations of the

envelope are inversely propor tional to frequency o:

¼ k

ð5Þ

We can generate multiple Gabor ﬁlt ers of different

sizes, orientation, or frequency by substituting differ-

ent parameters in Eq. 4. Gabor ﬁlters of different

orientation and frequency can capture different texture

details in iris texture. An example of Gabor ﬁlter as w ell

as Gaussian derivative ﬁlter are shown in Fig. 8.

We get a two-dimensional (2D) complex-valued

plane after applying each Gabor ﬁlter on iris texture.

Daugman proposed a

▶ phase-quadrant quantization

scheme to further quantize the 2D complex-valued

plane. Because amplitude information depends on ex-

traneous factors such as imaging contrast, illumina-

tion, and camera gain, it is not very discriminating. On

the contrary, phase information on a complex-valued

ﬁlter response plane has been shown to have much

more discriminative information in biometric recog-

nition [7, 11]. Therefore, performing phase quantiza-

tion would discard useless information and at the same

time, make feature representation more compact and

easy to handle.

The phase-quadrant quantization scheme amounts

to a patch-wise phase quantization of the iris pattern, by

identifying in which quadrant of the complex plane each

resultant phasor lies. For every location on the complex

plane, the sign of real and imaginary part jointly deﬁnes

in which quadrant the phasor lies. Two bitcodes are used

to encode the four possible quadrants. The phase-quad-

rant quantization scheme is illustrated in Fig. 9.

The phase-quadrant demodulation process is re-

peated across all the plane of the response of the ﬁlters,

with many different wavelet sizes, frequencies, and

orientations to extract totally 2048 bits. Figure 10

shows the raw eye image, unwrapped iris texture in

polar coordinate, and the iris code after feature

extraction.

Besides Gabor ﬁlters, there are also other ty pe of 2D

ﬁlters that can be used in iris feature extraction. Zhu

Iris Recognition, Overview. Figure 8 Examples of two filters which can be used for iris feature extraction. (a) Directional

Gaussian derivative ﬁlter, in 2D view (b) Directional Gaussian derivative ﬁlter, in 3D view (c) Gabor ﬁlter, in 2D view

(d) Gabor ﬁlter, in 3D view.

Iris Recognition, Overview. Figure 9 Illustration of

Daugman’s phase-quadrant demodulation scheme.

According to the quadrant each complex number falls in,

the angle of the phasor is quantized to one of the four

quadrants, setting two bits of phase information. This

process is repeated across all the iris with many wavelet

sizes, frequencies, and orientations.

816

Iris Recognition, Overview

et al. proposed to use 2D wavelet transform to perform

feature extr action [12]. By applying 2D wavelet trans-

form on an iris texture map, one can get a set of sub-

images of different resolution levels. They proposed to

use the mean and variance (or standard deviation) of

each wavelet sub-images to be the features.

Ma et al. proposed to use circular symmetric ﬁlters

to perform iris feature extraction [13, 14]. The kernels

of circular symmetric ﬁlters can be deﬁned as follows:

Gðx; y; f Þ¼

2pd

exp 

!"#

Mðx; y; f Þ

Mðx; y; f Þ¼cos 2pf

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

þ y

hi

; ð6Þ

where f is the frequency of the modulating sinusoidal

function, d

and d

are the space constants of the

Gaussian envelope along the x and y axis, respectively.

The difference between circular symmetric ﬁlters and

Gabor ﬁlters is that Gabor ﬁlter of a ﬁxed parameter

focus only on information obtained from a speciﬁc

orientation; w hile circular sy mmetric ﬁlters extract

information from all orientation. Circular symmetric

ﬁlters can also put more emphasis on a par ticular

orientation, by changing the parameter d

and d

After iris texture has been convolved with circular

symmetric ﬁlters, they proposed to extract the mean

and the average absolute deviation of the magnitude of

local patches (of size 8  8) of the convolved image as

feature vectors. Finally, Fisher line ar Discriminant

analysis was applied to reduce the dimensionality of

the feature vectors.

One last thing to note is that, very often, iris texture

is occluded by other objects, for example, eye lashes,

eye lids, or specular reﬂection from eye glasses, it is not

that every point on the iris texture map is useful for

pattern matching. Therefore, it is necessary to compute

an iris mask, which is exactly the same size as iris

texture, to indicate which part of the map is really

iris texture, and which part is not. This mask will be

used in the stage of pattern matching, as described in

the next section.

Pattern Matching

After iris code and its mask have been computed, the

next step is to match two irises to see if they are coming

from the same class. Obviously, to complete the task,

the only thing one has to do is to count how many bits

of these two irises share the same value. One thing to

note is that when we compare the differences of each

bit, we must not count the bit which belongs to the

region that is not iris texture. Therefore, we have to

mask out non-iris region before we compute the

matching score.

Another thing to note is that since we already

quantize the feature of iris texture into binary values,

there are only two possible values for each location on

iris map (1 or 0). There exists one operation which can

be executed extremely fast to compare if 2 bits share

the same value, which is the exclusive-or operation

(XOR). XOR returns zeros if both bits are the same

(both of them are zeros or ones), ones if they are

different (one of them is 0 and the other is 1). There-

fore, by performing bit-wise XOR operation, one can

compute the distance score for two irises in extremely

high efﬁciency.

To summarize, suppose we want to compare the

distance be tween iris A and B, whose two phase code

bit vectors are denoted {code_A, code_B} and whose

Iris Recognition, Overview. Figure 10 (a) Eye image with segmentation in Cartesian coordinate (b)

unwrapped iris image in polar coordinate (c) iris code computed by applying Gabor ﬁlters on (b) then perform

bit-wise quantization.

Iris Recognition, Overview

817

mask bit vectors are denoted {mask_A, mask_B}, it can

be computed simply by following equation:

HD ¼

ðcode A  code BÞ\mask A \ mask B

mask

A \ mask B

;

ð7Þ

where

denotes bit-wise_ XOR operation, \ denotes

bit-wise AND operation, and 

denotes the norms of

the resultant bit.

The computed score is a metric of ‘‘distance,’’

indicating how different these two irises are. It is

called ‘‘Hamming Distance’’ (HD). As one can see

from Eq. 7, Hamming Distance is normalized by the

norm of the effective region. Therefore, it will always

have values between 0 and 1, where 0 indicates if iris

A and B are exactly the same and 1 indicates if they are

completely opposite.

Ideally, if two irises are from the same class, the

Hamming Distance between them should be close to 0.

On the other hand, if two irises are from different

classes, due to the property of statistical independence,

the probability for each bit of one iris to match the

same bit of another iris should be 50%. Therefore, the

expected Hamming Distance for two irises which come

from different classes should be 0.5. In practice, the

values of Hamming Distance will not be exactly 0 or

0.5. The distribution of the value of Hamming Dis-

tance for authentic comparison will be a Gaussian

distribution, centered at 0, and the distribution of the

value of Hamming Distance for imposte r comparison

will be another Gaussian distribution, centered at 0.5.

If the qualit y of the input iris image is high enough, in

most cases, these two Gaussians would not intersect

with each other, or they only intersect with each other

in tiny portion. Therefore, a proper threshold can be

chosen to minimize both the

▶ False Positive Rate and

▶ False Negative Rate.

Summary

Iris recognition is an emerging ﬁeld for biometric

recognition. Substantial research effor ts have been

involved in this ﬁeld to push the performance of iris

recognition to the limit. Literature has shown that it is

one of the biometric modality that has high perfor-

mance, high universality, high distinctiveness, high

permanence, and low chances of circumvention [15].

As the technological innovation of iris acquisition

keeps advancing, it is becoming more user-friendly

and more popular. As the computational power of

hardware grows exponentially and the size of chip

keeps decreasing, iris segmentation, feature extraction,

and matching can all be executed faster in much smal-

ler devices. In the near future, it is very promising that

iris recognition system will be widely accepted, not

only in the application at national security level, but

also in private companies, public services, and private

residency.

Related Entries

▶ Biometrics Overview

▶ Biometric System Design

▶ Iris Acquisition Device

▶ Iris Encoding and Recognition

References

1. Flom, et al.: Iris recognition system. US Patent 4641349, 3

Feb 1987

2. Daugman, J.: Biometric personal identiﬁcation system based on

iris analysis. US Patent 5291560, 1 Mar 1994

3. Wildes, et al.: Automated non-invasive iris recognition system

and method. US Patent 5572596, 5 Nov 1996

4. Dowski, E.R. Jr., Johnson, G.E.: Wavefront coding: a modern

method of achieving high-performance and/or low-cost imaging

systems. Proc. SPIE 3779, 137 (1999), DOI:10.1117/12.368203

5. Wildes, R.: Iris recognition: An emerging biometric technology.

Proc. IEEE 85, 1348–1363 (1997)

6. Daugman, J.: High conﬁdence visual recognition of persons by a

test of statistical independence. IEEE Trans. Pattern Anal. Mach.

Intell. 15(11), 1148–1161 (1993)

7. Daugman, J.: How iris recognition works. IEEE Trans. Circ. Syst.

video Technol. 14(1), 21–30 (2004)

8. Daugman, J.: New methods in iris recognition. IEEE Trans. Syst.

Man Cybern. Part B, 37(5), 1167–1175 (2007)

9. Daugman, J.: Statistical richness of visual phase information:

Update on recognizing persons by iris patterns. Int. J. Comput.

Vis. 45(1), 25–38 (2001)

10. Daugman, J.: The importance of being random: Statistical prin-

ciples of iris recognition. Pattern Recognit. 36(2), 279–291

(2003)

11. Savvides, M., Kumar, B.V.K.V., Khosla, P.K.: Eigenphases vs

eigenfaces. In: Proceedings of the 17th International Conference

on Pattern Recognition 2004 (ICPR 2004), vol. 3, pp. 810–813

vol. 3, 23–26 (Aug 2004)

12. Zhu, Y., Tan, T., Whang, Y.: Biometric personal identiﬁcation

based on iris patterns. In: Proceedings of International Confer-

ence on Pattern Recognition, pp. 2801–2804. Barcelona, Spain

(2000)

818

Iris Recognition, Overview

13. Ma, L., Tan, T., Wang, Y., Zhang, D.: Personal identiﬁcation

based on iris texture analysis. IEEE Trans. Pattern Anal. Mach.

Intell. 25(12), 1519–1533 (2003)

14. Ma, L., Wang, Y., Tan, T.: Iris recognition using circular sym-

metric ﬁlters. In: Proceedings of International Conference on

Pattern Recognition, pp. 414–417 (2002)

15. Jain, A.K., Ross, A., Prabhakar, S.: An introduction to biometric

recognition. IEEE Trans. Circ. Syst. Video Technol. 14(1), 4–20

(2004)

Iris Recognition Algorithms

An iris recognition algorithm is a method of matching

an iris image to a collection of iris images that exist in a

database. There are many iris recognition algorithms

that employ different mathematical ways to perform

recognition. Breakthrough work by John Daugman

led to the most popular algorithm based on Gabor

wavelets.

▶ Iris Acquisition Device

Iris Recognition at Airports and

Border-Crossings

JOHN DAUGMAN

Computer Laborato ry University of Cambridge,

Cambridge, UK

Synonyms

CANPASS; CLEAR; Iris recognition immigration sys-

tem (IRIS); NEXUS; Privium; RAIC

Definition

As case illustrations of generic biometric applications,

there are at least ﬁve different modes in which auto-

mated personal identiﬁcation by iris recognition is

used at airports: (1) international arriving passengers

can clear Immigration control at iris-automated gates

without passport or other identity assertion if they

have been enrolled in a preapproved iris database;

(2) departing passengers can receive expedited security

screening and check-in as low-risk travelers if enrolled

in an iris database following background checks;

(3) airline crew members use iris recognition for con-

trolled access to the secure air-side; (4) airport employ-

ees gain access to restricted areas within airports such

as maintenance facilities, baggage handling, and the

tarmac; and (5) arriving passengers may be screened

against a watch-list database recording the irises of

persons deemed dangerous, or of expellees excluded

from entering a country. All such existing programs

use the Daug man algorithms for iris encoding and

recognition because of the need to process iris images

fully at the speed of the vid eo frame rate (30frames/s)

and to search databases at speeds of about a million

IrisCodes per second, and the need for robustness

against making False Matches in large databas e

searches despite so many opportunities. However,

the threat models posed for the different applications

are distinctive, depending on whether an attacker’s

goal is a False non-Match (a concealment attack, e.g.,

in a watch-list application) or a False Match (an

impersonation attack, e.g., to be taken for a registered

traveler in an expedited Immigration control or

trusted-traveler deployment). Likewise, the business

models vary for these different uses, depending

on whether the traveler pays for the convenience of

expedited processing, or an airport owner pays for

the facility’s enhanced security and productivity, or a

government funds such a technology deployment both

to improve process efﬁciency and to achieve national

security goals.

Introduction

Most deployments of biometric systems have as their

main purpose either enhancing the security and reliabil-

ity, or enhancing the convenience and efﬁciency, of an

identiﬁcation process. In some applications either secu-

rity or efﬁciency dominates the requirement, while the

other is less important. For example, in identifying

theme park visitors biometrically in lieu of ticketing,

efﬁciency is much more important than security; where-

as for biometric applications within prisons or detention

centers, just the opposite is the case. In airports, how-

ever, both of these objectives are paramount, and neither

can be compromised. Excelling simultaneously at both

Iris Recognition at Airports and Border-Crossings

819