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

Подождите немного. Документ загружается.

are also a number of minor differences. These are

summarized in Table 1.

Iris Standards Adoption

A number of national and international governments

and organizations have ofﬁcially adopted iris image

data standards for current and planned programs.

Within the US Department of Homeland Security the

Transportation Security Agency (TSA) has adopted

ISO/IEC 19794-6, in addition to ANSI INCITS stan-

dards for Finger Minutiae, Face Recognition, and the

Common Biometric Exchange Formats Framework

(CBEFF) for its Registered Traveler program. In this

program the iris rectilinear format is used for trans-

mission of enrollment images to a central data center,

and polar format is used for storage of iris image data

on the registered tr aveler card. The International Civil

Aviation Organization (ICAO) has adopted ISO/IEC

standards for face, ﬁnger, and iris biometrics, in addi-

tion to the ISO/IEC CBEFF standard, for its Machine

Readable Travel Documents (MRTDs).

Current Iris Standards Activity

Current standards activities related to the iris image

standards include development of conformance

test standards, development of rev isions of the current

standards, and adoption of the international ISO/IEC

standard by various national standards organizations.

Conformance testing is intended to assess commer-

cial products that claim to support the iris standard.

Developers of software applications and biometric

products may interpret the standard differently from

one another, and as a result their implementations of

the speciﬁcation may differ and not interoperate. Con-

formance to the standard is a necessary prerequisite for

achieving interoperability among implementations;

therefore, there is a need for a standardized, gener-

ally accepted, conformance testing methodology that

would allow implementation of a set of test tools

realizing this methodology. The conformance test stan-

dards are applicable to the development and use of

conformity test method speciﬁcations, conformity

test suites for Iris Image Data Record requirements as

speciﬁed the ANSI and ISO standards, and confor-

mance testing programs for conforming products.

They are intended primarily for use by testing organi-

zations, but may be applied by developers and users

of test method speciﬁcations and test method

implementations.

The international biometric standards committee

ISO/IEC JTC1 SC37 is currently developing a revised

version of ISO/IEC 19794-6 to address needed clariﬁ-

cation of the original standard, incorporate certain

technolog y innovations, and respond to new customer

requirements. In particular, SC37 is considering both

removal of the requirement for mandatory embedding

of the iris image record within a CBEFF data structure

and elimination of the polar image format.

International standards such as ISO/IEC 19794-6

may be adopted by national standards bodies for use as

national standards. The U.S. has adopted the interna-

tional iris standard as a US standard, and has with-

drawn ANSI INCITS 379, cancelling the revision

project for this standard and the conformance testing

methodolog y standard project.

Summary

The published ANSI INCITS and ISO/IEC standards for

iris image data interchange were developed in a cooper-

ative, collaborative process with participants from nu-

merous commercial entities, government agencies, and

Iris Image Data Interchange Formats, Standardization. Table 1 Comparison of ANSI and ISO versions of iris image

interchange standards

Attribute ANSI INCITS 379 ISO/IEC 19794-6

Image quality field

description

Described as 4 categories with numerical range

1–100

Described as value with numerical range

1–100

Second level header title Feature Header Biometric Subtype Header

CBEFF Product Identifier Contained in Iris Record Header Contained in CBEFF Header

User Identification No. Contained in Iris Record Header Contained in CBEFF Header

800

Iris Image Data Interchange Formats, Standardization

academic institutions from the US and other countries.

The standards have been successfully adopted by a num-

ber of US and international organizations that desire to

use iris recognition in operational deployments. The

standards are sufﬁciently ﬂexible to accommodate the

products of diverse commercial vendors and user orga-

nizations. Ongoing standards development work will

result in more ﬂexible and interoperable standards and

conformance test suites that may be used to effectively

test the compatibility of various products with the

standards.

Related Entries

▶ Biometric Sample Quality, Standardization

▶ Biometric Technical Interface, Standardization

▶ Common Biometric Exchange Framework Formats,

Standardization

▶ Conformance Testing Methodologies for Biometric

Data Interchange Formats, Standardization of

▶ Data Interchange Format, Standardization

▶ Iris Image Quality

▶ Iris Recognition, Overview

References

1. Flom, L., Saﬁr, A.: Iris recognition system, US Patent

No. 4,641,349, United States Patent and Trademark Ofﬁce,

Washington DC, (1984)

2. Kim, D., Ryoo, J.: Iris identiﬁcation system and method

of identifying a person through iris recognition, US Patent No.

6,247,813, United States Patent and Trademark Ofﬁce, (2001)

3. Monro, D., Rakshit, S., Zhang.: DCT-Based Iris Recognition.

IEEE Trans. Pattern Anal. Mach. Intell. 29(4), 586–595 (2007)

4. Wildes, R., Asmuth, J., Hanna, K., Hsu, S., Kolczynski, R.,

Matey, J., McBride, S.: Automated, non-invasive iris recognition

system and method, US Patent No. 5,572,596, United States

Patent and Trademark Ofﬁce, (1996)

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

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

Intell. 15(11), 1148–1160, (1993)

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

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

Intell. 25(12), 1519–1533, (2003)

7. Masek, L.: Recognition of human iris patterns for biometric

identiﬁcation, Bachelor of Engineering thesis, School of Com-

puter Science and Software Engineering, The University of West-

ern Australia, (2003)

8. Tisse, C., Martin, L., Torres, L., Robert, M.: Person identiﬁcation

technique using iris recognition, In: Proceedings of the 15th

International Conference on Vision Interface, Calgary, Italy,

pp. 294–299 (May 29, 2006)

9. ANSI INCITS 379-2004 Iris image interchange format. Ameri-

can National Standards Institute, (2004)

10. ISO/IEC 19794-6: 2005 Information technology – Biometric

data interchange formats – Part 6: Iris image data, International

Standards Organization, (2005)

11. ISO/IEC 19785-1:2006 Information technology – Common bio-

metric exchange formats framework – Part 1: Data element

speciﬁcation, International Standards Organization, (2006)

Iris Image Enhancement by

Super-Resolution Method

▶ Iris Super-Resolution

Iris Image Quality

NATALIA A. S CHMID

West Virginia University, Morgantown, USA

Synonyms

Information content of iris images; Iris quality metrics

Definition

Iris image quality evaluation is a procedure of measur-

ing information content of iris imagery at the stage of

iris acquisition or at early processing stage. The infor-

mation content may be decided to be insufﬁcient to be

used for iris identiﬁcation based on a sing le image. In

this case, the image may be discarded, or combined

with other imagery to improve recognition capabilities

of an iris system. Evaluated quality metrics would be

the guidelines in making decisions regarding further

steps with respect to acquired imagery.

Introduction

Iris image quality assessment is an important research

thrust recently identiﬁed in the ﬁeld of iris biometrics

[1–3]. This research is tightly related to the research on

Iris Image Quality

801

▶ nonideal iris. Its major role is to determine, at the

stage of data acquisition or at the early stage of proces-

sing, what the amount of information for the purposes

of processing, recognition, and fusion this imagery

contains. Is it informative enoug h for performing fur-

ther processing steps or should be discarded? Is it

informative enough for being combined with other

images and result in improved recognition perfor-

mance? The quality metrics play an important role in

automated biometric systems for three reasons: (1)

system performance (segmentation and recognition),

(2) interoperability, and (3) data enhancement.

The quality metrics play an importan t role in auto-

mated biometric systems for two reasons: (1) system

performance (segmentation and recognition), and (2)

interoperability.

A traditional approach in evaluating iris imag e

quality is to identify a single or a sequence of physical

phenomena that inﬂuences formation of query iris

imagery at the image acquisition stage (see the entry

on Image Acquisition). The distortions that identiﬁed

physical phenomena introduced are then modeled

mathematically. To evaluate the level of distortion pres-

ent in an iris image, a single or a set of metrics or quality

measures is speciﬁed. The metrics can be absolute

measures or relative measures. The abs olute metrics

do not assume comparison of query image with a

reference image. The relative metric measures the pres-

ence of some distortions with respect to a speciﬁed

reference image.

The following sections provide a short survey of the

literature on iris image quality, introduce a set of iris

quality measures and suggest a number of techniques

to combine individual quality measures in a single

score.

Survey of Iris Quality Metrics

Previous work on iris image quality can be placed in tw o

categories: local and global analysis. Zhu et al., [4]evalu-

ate quality by analyzing the coefﬁcients of particular areas

of iris’s texture by employing discrete wavelet decompo-

sition.Chenetal.,[5] classify iris qu ality by measuring

the energy of concentric iris bands obtained from 2-D

wavelets. Ma et al., [6] analyze the Fourier spectra of

local iris regions to characterize out-of-focus and mo-

tion blur and occlusions. Zhang and Salganicaff [7]

examine the sharpness of the region between the pupil

and the iris. Daugman [8] and Kang and Park [9]

characterize quality by quantifying the energy of high

spatial frequencies over the entire image region.

Belcher and Du [10] propose a clarity measure by

comparing the sharpness loss within variou s iris

image regions against the blurred version of the same

regions. The major feature of these approaches is that

the evaluation of iris image quality is reduced to the

estimation of a single [5, 7, 8, 9] or a pair of factors [6],

such as out-of-focus blur, motion blur, and occlusion.

A broader range of physical phenomena that can be

observed in nonideal iris imagery was characterized

by Kalka et al., [11, 12]. The proposed factors include

out-of-focus and motion blur, occlusion, specular re-

ﬂection, illumination, off-angle, and pixel count. The

strength of the phenomena and its inﬂuence was eval-

uated through modiﬁed or newly designed iris quality

metrics. These factors based on the extensive analysis

carried out by the authors affect the segmentation and

ultimately recognition performance of iris recognition

systems. An example of two iris images from ICE

dataset [13] and their corresponding pentagram plots

are displayed in Figure 1 and Figure 2. In a pentagram ,

each axis represents a quality metric. The quality score

is normalized to take values between zero and one. The

value one corresponds to the lowest quality, the value

zero is the highest quality.

Since most of quality metrics contain some com-

mon information (for example, motion and out-of-

focus blurs are physically related; occlusion and pixel

counts usually contain redundant information; illumi-

nation and contrast are also related), they have to be

estimated jointly. However, all recently designed qua-

lity assessing algorithms treat individual quality metrics

as independent and thus evaluate them separately.

While any processing reduces information content

of quality metrics, in some applications it is beneﬁcial

to have a single quality score that characterizes the

overall image quality. In this case we would appeal to

rules of combining scores.

Rules of Combination

The quality factors (metrics) can be used individually

or combined into a single score throug h a simple

static or an adaptive rule. Among static rules the sim-

ple sum rule is a computationally efﬁcient method .

More complex (adaptive) rules such as Bayesian,

802

Iris Image Quality

Dempster-Shafer, weighted Sum, or any previously

designed fusion strategy to combine classiﬁers can

also be used to combine quality metrics into a single

score. These rules are more fundamental and ﬂexible,

but require intensive computations.

Relevance to Recognition

Performance

It is well understood that the capabilities of various

designed metrics to evaluate the quality of iris image s

has to be related to their capabilities to predict recog-

nition performance by quickly analyzing the quality of

imagery. Since in most analyzed cases recognition per-

formance is nonlinearly related to quality metrics, it is

hard to evaluate precision with which quality metrics

and recognition performance are related. It is because

of this reason most of the designed quality metrics are

not highly correlated with one another.

Figure 3 demonstrates the relationship between

the iris quality metrics evaluated by following the

procedure in Kalka et al., and combined through app-

lication of Dempster-Shafer rule and the recognition

performance. The results are obtained using images

from ICE dataset [13]. To obtain the plots in Figure 3,

each image undergoes quality evaluation. The images

are ranked based on their quality score. The receiver

operating curves are plotted for three cases. In the ﬁrst

case, the entire dataset is used. In the second case, the

images with the quality above the score 0.75 are select-

ed. In the last case, the images with the quality above

the score 0.85 are selected. Note that here the score

value one corresponds to the highest quality. From

Figure 3, the combined metric predicts recognition

performance relatively well.

This indicates that the information extracted fro-

miris imagery is correlated to some degree with the

recognition performance. The question that remains to

be addresses if it is possible to establish an explicit

Iris Image Quality. Figure 2 Pentagram plot of quality for the images in Figure 1.

Iris Image Quality. Figure 1 Sample images from ICE dataset. (a) Good quality image. (b) Poor quality image.

Iris Image Quality

803

relationship between quality metrics and recognition

(veriﬁcation) performance.

In practice most of metrics are designed to me asure

signal-to-noise ratio, ratio of powers, directional

power, sharpness of edges, ratio of pixels, empirical

entropy and relative entropy. These metrics are not

explicitly related to probability of recognition error

or receiver operating characteristic (ROC) curve, two

traditional measures of recognition and veriﬁcation

performance. Furthermore, the degree of nonlinearity

and nonlinear model relating the quality metrics and

recognition performance are hard to evaluate.

The conclusions above are supported by the results

of an extensive research published in the literature on

the topic of feature selection. The problem of feature

selection can be brieﬂy summarized as follows. Due to

suboptimality of an image/signal encoding procedure

(feature extraction procedure) templates contain a cer-

tain number of noisy components or components un-

informative for pattern recognition. Removal of these

components, therefore, often leads to improved recog-

nition performance. In the past, some substantial

efforts have be en made t o ﬁnd an information measure

that would be capable of evaluating the information

content of a feature and, at the same time, would be

able to predict changes in recognition performance due

to removal of a feature [14, 15]. In spite of long term

efforts, the ﬁnal conclusions may not look worth of

these efforts: (1) there is a single family of statistical

models (Gaussian family) that allows establishing an

explicit (exponential) relationship between the infor-

mation content of a feature and the recognition per-

formance. Unfortunately for biometrics, only few

types of biometric templates (encoded data) can be

modeled as being Gaussian distributed. (2) The tradi-

tional measures of infor mation suc h as signal-to-

noise-ratio, entropy, relative entropy, and mutual

information, are only asymptotically related to proba-

bility of recognition error [16]. When these measures

are adapted to the problem of empirical evaluation of

the information content of a feature, the relationship

between the empirically evaluated measures and the

probability of recognition error can be established only

under the condition that an increasing amount of data

can be used to evaluate information in features and

provided that the empirical information measures

converge in some sense to the true measures.

Summary

Iris images quality metrics provide us with a fast

way to predict information content of biometric data.

Current quality metrics are designed to evaluate qua-

lity factors individually, that is, separately. These quality

factors may further be combined by invoking static

or adaptive rules applied to individual scores. Since

various quality factors contain redundant infor-

mation about one another, a more optimal solution

to the problem of evaluation of iris image quality

would be to evaluate the quality factors jointly. Also,

identifying the most important quality factors that

inﬂuence the recognition performance remains an

open problem.

Related Entries

▶ Iris Segmentation

▶ Iris Recognition

▶ Score Fusion and Normalization Rules

References

1. Biometric quality workshop. national institute of standards and

technology. (2006)

2. Biometric quality workshop ii. national institute of standards

and technology. (2007)

3. Bowyer, K., Hollingsworth, K. Flynn, P.: Image understanding

for iris biometrics: A survey. J Comp Vision and Image under-

standing. 110, 281–307 (2007)

Iris Image Quality. Figure 3 Establishing the relationship

between estimated quality metrics and verification

performance of an iris recognition system. The results are

based on applications of a Gabor-filter based encoding

algorithm to iris images from ICE dataset.

804

Iris Image Quality

4. Zhu, X., Liu, Y., Ming, X., Cui, Q.: A quality evaluation method

of iris images sequence based on wavelet coefﬁcients in region of

interest. In: Proceedings of the fourth International Conference

on Computer and Information Technology, pp. 24–27 (2004)

5. Chen, Y., Dass, S., J.A.: Localized iris quality using 2-d wavelets.

In: Proceedings of International Conference on Biometrics,

pp. 373–381. Baltimore, MD (2006)

6. Ma, L., Tan, T., Zhang, Y.W.D.: Personal identiﬁcation based on

iris texture analysis. IEEE Trans. Pattern Anal. Mach. Intell.

25(12), 1519–1533 (2003)

7. Zhang, G., Salganicoff, M.: Method of measuring the focus of

close-up image of eyes (1999)

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

Video Technol. 14(1), 21–30 (2004)

9. Kang, B., Park, K.R.: A study on iris image restoration. In:

Proceedings Audio Video Based Personal Authent., vol. 3546

(2005)

10. Belcher, C., Du, Y.: Information distance based selective feature

clarity measure for iris recognition. In: Proceedings SPIE

Symposium on Defense and Security. Conference on Human

Identiﬁcation Technology IV., vol. 6494 (2007)

11. Kalka, N.D.: Image quality assessment for iris biometric. https://

eidr.wvu.edu/ﬁles/4447/Kalka_Nathan_thesis.pdf (2005)

12. Kalka, N., Zuo, J., Dorairaj, V., Schmid, N., Cukic, B.: Image

quality assessment for iris biometric. In: Proceedings of 2006

SPIE Conference on Biometric Technology for Human Identiﬁ-

cation III. Orlando, FL (2006)

13. Liu, X., Bowyer, K.W., Flynn, P.J.: Iris recognition and veriﬁca-

tion experiments with improved segmentation method. In: pro-

ceedings Fourth IEEE Workshop on Automatic Identiﬁcation

Advanced Technologies (AutoID) (2005)

14. Ben-Bassat, M.: Use of distance measures, information measures

and error bounds in feature evaluation. Handbook of Statisti-

cal, P.R. Krishnaiah and L.N. Kanal, North-Holland Publishing

Company, pp. 773–791 (1982)

15. Jain, A.K., R.P.W.D., Muo, J.: Statistical pattern recognition: A

review. IEEE Trans on Pattern Analysis and Machine Intelli-

gence. 22, 4–36 (2000)

16. Cover, T.H., Thomas, J.A.: Elements of Information Theory.

Wileg Series in Telecommunications, New York (1991)

Iris Interchange Format Standards

▶ Iris Image Data Interchange Formats ,

Standardization

Iris Localization

Iris localization is one important stage in iris recogni-

tion system. The goal of iris localization is to ﬁnd the

boundary between iris and sclera and between iris and

pupil. Because the shape of pupil and the shape of iris

look very close to a circle, most of the iris localization

algorithm tries to perform circle ﬁtting on the eye

image. In this way, the results of iris localization can

be expressed as two circles.

▶ Automatic Class ification of Left/Right Iris Image

▶ Iris Super-Resolution

Iris on the Move™

JAMES R. MATEY

Electrical and Computer Engineering Department,

US Naval Academy, Annapolis, MD, USA

Synonyms

Drive-up; Iris at a glance; Minimal constraint iris

recognition; Portal; Walk-through; Walk-up

Definition

Iris on the Move™, commonly referred to as IOM, is

an approach to acquiring iris images suitable for iris

recognition while minimizing the constraints that need

to be placed on the subject. Sarnoff Corporation

developed IOM under contract to the United States

Government: NMA401-02-9-2001.

Multiple implementations of IOM have been

demonstrated; they include: a portal walk-through

system, an on-the-wall walk-up system, an over-the-

door walk-through system and a roadside drive-up

system. The key features of the IOM systems are large

▶ capture volume, large ▶ standoff distances and the

capability of capturing recognition quality iris images

while the subject is in motion [1].

The Panasonic ‘‘Iris at a Glance’’ systems and

the H-box™ from the Hoyos Group are examples of

related technologies.

Introduction

Iris recognition is one of the strongest biometric avail-

able [2–4]. Iris recognition is a strong biometric

Iris on the Move™

805

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 – in the absence of illness or injury that

disrupts the iris t issue.

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 1990s 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 2007, though there are

vigorous research efforts into alternative algorithms

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

rithms in current use is iris2pi.

The capability of providing extremely low

▶ false

match rates is one of the great strengths of iris2pi.

Daugman has published credible evidence that iris2pi

can support false match rates of the order of a part in a

trillion [9]. However, the quality of the iris images

required for such results is high. The standards for

iris images published by the ISO [10] call for 100–

200 pixels across the iris and 40 dB signal-to-noise -

ratio (SNR). In the decade following the development

of the Daug man algorithm, Iridian, IrisGuard, LG,

OKI, Panasonic, and Securimetrics all brought iris

recognition cameras to market that could provide

such high quality iris images. Figure 1 is an example

of one of these cameras that provides the full 200 pixels

across the iris.

All these systems require substantial cooperation

from the subjects and impose signiﬁcant constraints

on them. In general, imposing constraints and requir-

ing cooperation reduces the ease of use of any biomet-

ric system. This may not be a signiﬁcant issue for

▶ habituated subjects, but it is a concern for nonhabi-

tuated subjects – subjects who do not use the system

on a routine basis.

Iris on the Move™ was developed to relax subject

constraints and make it possible for subjects to use an

iris recognition system with little instruction. The con-

straints of greatest interest are capture volume, stand-

off,

▶ residence time, subject motion, effects of

ambient illumination, and overall ease of use. Capture

volume is the three dimensional volume through out

which the iris camera is capable of acquiring an

acceptable image. Standoff is the distance from the

camera to the subject; in some systems there may be

two standoff distances – subject to camera and subject

to illumination. Residence time is the length of time

that a subject must stay with in the capture volume to

enable collection of an acceptable image. Subject mo-

tion within the capture volume can be characterized in

terms of a velocity – the maximum velocity at which a

subject can move during collection of an acceptable

image. Overall ease of use can be characterized in terms

of the complexity of the instructions that must be

provided to an uninitiated subject to insure acceptable

iris image collection on their ﬁrst use of a system.

The reader may well ask, ‘‘What is an acceptable iris

image?’’. Operationally an acceptable image is a recog-

nition image that matches well against an enrollment

image taken under ideal circumstances – the enrollment

image meets ISO standards, highly constrained subject,

no time limit, repeated trials available to obtain opti-

mum quality image, and close supervision of the enroll-

ment process by a well trained operator. It is important

to note that iris2pi looks at the local phase of the

iris image in predeﬁned spatial frequency bands (see

the entry on Gabor wavelet iris encoding). Hence, the

Iris on the Move™. Figure 1 An enrollment quality iris

image capture device, the LG IrisAccess 3000. (Photo

provided courtesy of Mohammed Murad of LG Electronics

USA Inc., Iris Technology Division.)

806

Iris on the Move™

image quality metrics that are used for assessing the

visual quality of images in daily life do not map directly

to the quality metrics for an image used for iris recogni-

tionwith iris2pi. For example, a reduction in the contrast

of a properly adjusted television display will almost

always reduce its visual quality – while a reduction in

the contrast of a good quality iris image may have little

impact on its match score against another good image of

the same iris. This can be understood in mathematical

terms: phase and amplitude are orthogonal/independent

coordinates. Human vision is more sensitive to ampli-

tude than phase; iris2pi is more sensitive to phase than

amplitude.

IOM Design Considerations

The single most important IOM design consideration

is the exploitation of asymmetries between enrollment

and recognition. As Daugman points out [9], the

number of degrees of freedom (as measured by the

number of valid bits in an iris2pi template) in an iris

template can vary depending on the quality of under-

lying image. Fewer degrees of freedom translate to a

broader

▶ imposter distribution. This broadening is

taken into account in many systems by adjusting the

Hamming distance of a match based on the number

valid bits compared between two templates to maintain

a constant false match rate for a ﬁxed match criterion in

the presence of such variation. At enrollment, signiﬁcant

effort can be expended, once per subject, to capture a

pristine image of the iris. Pristine enrollment images

enable us to purge the database of duplicates – even

when the database is large. They also ensure that maxi-

mal use is made of the degrees of freedom available in a

lower quality recognition image.

If there is high quality enrollment, at recognition,

it can afford a reduction in the quality of the image.

The

▶ authentic s distribution will be dominated by

the quality of the recognition images. Reduction

in the quality of the recognition images will broaden

the authentics distribution and increase the

▶ false

non-match rate (FNMR). However, it is possible to

adjust for the increase in the single attempt FNMR

through use of multiple attempts.

IOM systems use high quality images from iris

cameras equivalent to that in Fig. 1 to build a strong

iris database. At each recognition attempt IOM sys-

tems take multiple, lower quality images of the subject

iris and compare them a gainst the database. Let FMR

(1) be the false match rate for a recognition attempt

where only one image is collected, and FNMR(1)

corresponding false non-match rate. Ignoring correla-

tions and some other statistical niceties, the ﬁrst

order effect of testing N images at each recognition

attempt is to change the performance of the system to

FMR(N)  N*FMR(1) and FNMR(N)  FNMR(1)

The important point is that FMR(N) increases slowly

(linearly) with N, whereas the FNMR(N) decreases

much more rapidly (exponentially).

Current implementations of iris2pi can generally

provide a better FMR than is necessary in most recog-

nition applications. Hence, a slightly higher FMR(N)

can be a trade-off for a much lower FNMR(N). This

enables us to use lower quality recognition images.

Lower quality on the recognition image can then

be converted into larger standoffs and larger capture

volumes for a ﬁxed system cost. IOM systems can be

designed to operate at the bottom end of the image

quality standards set by the ISO. In particular

the systems can be designed for 100 pixels across

the iris and somewhat less than 40 dB SNR.

The essay now considers each of the system para-

meters in turn and discusses the relevant tradeoffs.

Capture Volume

Capture volume is the product of two factors: ﬁeld of

view (an area) and

▶ depth of ﬁeld (a distance). The

ﬁeld of view is product of the width and heig ht of

the region in focus for the imager. Field of view is

determined by the pixel count of the camera sensor.

A 2048  2048 pixel camera will give a ﬁeld of view of

20.48  20.48 cm at a subject iris resolution of 100

pixels/cm. IOM systems achieve a large ﬁeld of view by

accepting an iris resolution of only 100 pixels/cm and

by using cameras with high pixel counts.

Depth of ﬁeld is the distance along the axis con-

necting the subject and the camera over which the iris

is in focus – without additional adjustment of the lens.

Depth of ﬁeld is a characteristic of the lens system and

is discussed in detail elsewhere [11]. Depth of ﬁeld

increases with increasing F# of the lens, which in turn

reduces the amount of light that gets through the lens

and the SNR of the image. IOM systems achieve effec-

tive depth of ﬁelds of the order of 10 cm at standoffs of

2–3 m by providing sufﬁciently bright illumination

to enable use of higher F#’s while still maintaining a

good signal-to-noise-ratio.

Iris on the Move™

807

Standoff

Once the pixel size of the imager and subject resolution

are chosen, the magniﬁcation of the lens system is

determined by the rules of geometrical optics. The

magniﬁcation combined with the camera standoff

determines the focal length of the lens through the

lens equation 1/f = 1/p + 1/q where f is the focal length,

p is the subject distance, and q is the image distance.

The magniﬁcation links p and q: M = q/p.

IOM systems achieve camera standoffs of 2 m or

more by using long focal length lenses (200 mm)

with F#’s determined by

▶ diffractio n limits and

depth of ﬁeld considerations.

Iris recognition systems almost always use some

form of active near IR illumination. Bot h, the illumi-

nation a nd the camera have standoffs. IOM systems

achieve illumination standoffs of 1–2 m by using arrays

of powerful near IR LEDs to deliver irradiance of the

order of 2 mW/cm

to the subject. The irradiance at

the subject is limited by safety considerations. The

▶ threshold limit valu es (TLV) for near IR irradiati on

are published by the ACGIH [12]. The TLVs limit both

the subject irradiance (W/cm

) and the source radi-

ance (W/cm

-sr), which is a characteristic of the LEDs

used in the illuminator.

Residence Time

The product of residence time and the image capture

rate (frame rate) of the camera gives the number of

images captured during a recognition attempt. This

must exceed one – preferably more than one to get

advantage from the multiple attempt tradeoff de-

scribed above.

IOM system use relatively high frame rate cameras

(15–60 fps) to minimize residence time while still

getting enough images to reliably perform recognition.

Subject Motion

Subject motion can be divided into two types: longitu-

dinal motion, along the camera axis, and transverse

motion, perpendicular to the camera axis. Subject mo-

tion has two primary effects on image capture: (1) it

limits the residence time within the capture volume

and (2) it introduces motion blur, which effectively

degrades the image resolution.

If the capture volume is 10 cm thick – set by the

depth of ﬁeld – and the subject is walking through the

volume with a longitudinal velocity of 1 m/s, he will

transit the volume in 0.1 s; his residence time is 0.1 s.

The product of his residence time and the image cap-

ture rate must be greater than 1, as noted above.

A rough measure of motion blur is the amount of

motion, in pixels, that occurs during the acquisition of

an image. To maintain the resolution of the system this

needs to be less than 1 pixel. Longitudinal velocity, v

has a blurring effect that is the result of magniﬁcation

change as the subject approaches the camera. The size

of the change is approximately the ratio of the distance

moved to the camera standoff. Since the iris is 100

pixels across, the magniﬁcation change must be kept

below 0.01. If the shutter time of the camera is t, and

the camera standoff is d

, then 0.01 > v

t /d

. For a

1 ms shutter, and 2 m standoff, the longitudinal veloc-

ity must be less than 20 m/s – about twice as fast at the

world record (10 s) in the 100 m dash. Hence, the

maximum longitudinal velocity is set by required resi-

dence time, rather than motion blur.

Transverse velocity is a much bigger issue. Motion

perpendicular to the camera subject axis simply moves

the pixels at the iris as much as the motion. For a tran-

sverse velocity v

, a shutter time, t, and a subject reso-

lution, d

(m/pixel) v

t < d

is required. For 100 pixels/

cm and a shutter time of 1 ms, v

< 0.1 m/s – much

smaller than the longitudinal velocity limit.

The IOM systems use shuttered cameras and

strobed illumination to freeze the subject motion.

However, 1 ms is, at present, a practical limit on the

strobes and shutters. The IOM systems use human

factors engineering to minimize the transverse motion

of the subject as they move toward the camera. The

subject is given a target to walk towards and is

instructed to walk in a straight line.

The strobe and shutter control the amount of light

that reaches the camera sensor. Shorter strobes/shut-

ters at the same peak irradiance reduce the number of

photons reaching the sensor and reduce the SNR. It is

possible to increase the light intensity during the strobe

to recover part of this, but there are limitations im-

posed by the maximum current that can be used to

drive the LEDs and maximum safe radiance and irra-

diance set by the TLVs.

808

Iris on the Move™

Ambient Light

Ambient light can interfere with iris recognition sys-

tems in various ways. Bright ambient lights can cause

extreme pupil constriction and can cause subjects to

squint; both of these effects can cause difﬁculty in

generating a good quality iris template from the

image. However, short of reducing the ambient, there

is little that can be done to ameliorate these physiolog-

ical responses. Bright ambient lights can also cause

specularities on the iris that interfere with the genera-

tion of iris templates. This factor can be ameliorated by

using narrow band illumination w ith narrow band

ﬁltering on the camera. It can also be ameliorated

with the strobed illumination and shuttered camera

used for blur reduction.

IOM systems reduce the effect of ambient illumi-

nation by using optical ﬁlters that block a large fraction

of the ambient light while passing the active illumina-

tion and by using strobed illumination and opening

the shutter only during the strobe, thereby rejecting the

ambient light that impinges on the subject during the

time the strobe is off.

Ease of Use

The IOM systems use human factors designed to max-

imize ease of use. People ﬁnd it relatively easy to walk

down the center of a portal, to walk along a line

painted on the ﬂoor, or to walk directly toward a

target from a given starting point. The ﬁelds of view

of the IOM systems have been chosen to insure that a

person carrying out such a task will, as a matter of

course, pass their eyes through the capture volume of

the system.

The IOM systems have been designed to take ad-

vantage of the motion of the subject toward the cam-

era, to avoid the need for the subjects to position

themselves at the focus of the system – they will walk

through it naturally as shown in Fig. 2. The instruc-

tions for a subject about to use an IOM portal for the

ﬁrst time are:

1. Open your eyes

2. Look straight ahead at your reﬂection in the camera

cover

3. Walk down the center of the portal at a moderate pace

4. Try to be recognized

System Approach

In the world of software engineering, a ‘‘spaghetti

design’’ is anathema – modularity is the mantra –

with each module distinct and independent. In the

IOM systems, photons thread essentially all of the

hardware components. Modularity, in one sense, is

not possible. Changes in the camera sensor will have

unavoidable repercussions for the illumination. The

key to designing an IOM system is to recognize the

need for system wide optimization.

Future Work

Researchers at numerous institutions are working to

expand capture volumes, extend standoff distances,

reduce residence times and decrease the cost of iris

recognition systems. They are also working to reduce

constraints on subject pose, a topic not covered here.

The development of higher resolution camera sensors

with higher sensitivity and higher frame rates and the

development of higher output near IR LEDs will

enable these improvements.

Alternative iris recognition algo rithms are also

being developed, as noted above. It is conceivable

that a new algorithm may enable interesting new tra-

deoffs in the design of iris cameras that will help us

further extend the ease of use of iris recognition.

Iris on the Move™. Figure 2 An Iris on the Move™ portal

system. (Photo provided courtesy of Sarnoff Corporation.)

Iris on the Move™

809