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

Подождите немного. Документ загружается.

matcher dependent. Since ﬁngerprint individuality is

assessed based on minutia e location and direction

only, the matcher of [10] was used which depends

only on minutiae information.

Note that as m or n or both increase, the values of

PRCs for both the models become large as it becomes

much easier to obtain spurious matches for larger m

and n values. Additionally, Table 3 illustrates an impor-

tant fact: The PRCs based on the mixture models are

orders of magnitude larger compared to Pankanti’s

model and closer to the empirical probability of at

least w matches. Note also that the mean of ls (the

theoretical mean number of matches) are closer to

the empirical counterpart (mean number of observed

matches) compared to Pankanti’s model. This demon-

strates the adequateness of the mixture models for

the assessment of ﬁngerprint individualit y. While the

mixture models is more adequate at representing min-

utiae variability, the PRCs obtained are far too large

indicating a large amount of uncertainty in declaring a

match between a ﬁngerprint pair. One way to reduce

the PRC is to add more ﬁngerprint features when

performing the identiﬁcation. Fingerprint individuali-

ty assessment can then be made by developing appro-

priate statistical models for these features.

Summary and Future Work

In this chapter, an overview of the challenges involved

in assessing the individualit y of ﬁngerprints is

presented. Two works have been discussed. Pankanti’s

model is the ﬁrst attempt at modeling the observed

minutiae distribution via statistical models, whereas

Zhu et al. developed more ﬂexible models that ade-

quately describe all minu tiae characteristics. There are

many open problems that still remain unsolved. Both

works have only addressed the issue of interclass min-

utiae variability. Appropriate statistical models for

modeling the intraclass minutiae variability are still

very few in the literature. A very important source of

intraclass variability is the quality of the query and

template images, and work still needs to be done to

investigate how PRCs change with quality of the ﬁn-

gerprint image. It is also important to develop statisti-

cal models for more complex ﬁngerprint features. In

this case, one can use more useful matching criteria

that utilize richer ﬁngerprint features.

In this chapter, the PRCs have been related to the

false match rates. Another measure of ﬁngerprint indi-

viduality should be related to the false nonmatch rates.

Eventually, a measure of ﬁngerprint individuality should

be a optimal combination of the two measures of errors.

Acknowledgments

The authors like to thank Prof. Anil Jain for introduction

of the ﬁngerprint individuality problem to them and

for many subsequent discussions that has helped them

in their research in this area. This chapter was written

under the support of the NSF DMS grant 0706385.

Individuality of Fingerprints. Table 3 A comparison between fingerprint individuality estimates using the (a) Poisson

and mixture models, and (b) Pankanti et al. [2]

Database (m,n,w)

Empirical Mixture Pankanti

Mean no. of matches PRC Mean l PRC Mean l PRC

NIST (52,52,12) 7.1 3.9  10

3

3.1 4.4  10

3

1.2 4.3  10

8

FVC2002 DB1 (51,51,12) 8.0 2.9  10

2

4.9 1.1  10

2

2.4 4.1  10

6

FVC2002 DB2 (63,63,12) 8.6 6.5  10

2

5.9 1.1  10

2

2.5 4.3  10

6

Individuality of Fingerprints. Table 2 Table giving the mean m and n in the overlapping area, the mean overlapping

area and the value of M for each database [7]

Database (m,n) Mean overlapping area (pixel

)

NIST (52,52) 112,840 413

FVC2002 DB1 (51,51) 71,000 259

FVC2002 DB2 (63,63) 110,470 405

750

Individuality of Fingerprints

Related Entries

▶ Fingerprint

▶ Fingerprint Individuality

▶ Fingerprint Matching, Automatic

▶ Fingerprint Matching, Manual

▶ Forensic Evidence of Individuality

References

1. Daubert v. Merrel Dow Pharmaceuticals Inc.: 509 U.S. 579,

113 S. Ct. 2786, 125, L.Ed.2d 469 (1993)

2. Pankanti, S., Prabhakar, S., Jain, A.K.: On the individuality of

ﬁngerprints. IEEE Trans. Pattern Anal. Mach. Intell. 24(8),

1010–1025 (2002)

3. United States v. Byron Mitchell: Criminal Action No. 96-407,

United States District Court for the Eastern District of Pennsyl-

vania (1999)

4. United States Department of Justice: Document SL000386,

March 2000. Online: http://www.forensic-evidence.com/site/

ID/IDfpValidation.html

5. Scolve, S.C.: The occurence of ﬁngerprint characteristics as a two

dimensional process. J. Am. Stat. Assoc. 74(367), 588–595

(1979)

6. Stoney, D.A., Thornton, J.I.: A critical analysis of quantitative

ﬁngerprint individuality models. J. Forensic Sci. 31(4),

1187–1216 (1986)

7. Zhu, Y., Dass, S.C., Jain, A.K.: Statistical models for assessing the

individuality of ﬁngerprints. IEEE Trans. Inf. Forensics Secur.

2(3), 391–401 (2007)

8. NIST: 8-bit gray scale images of ﬁngerprint image groups

(FIGS). Online: http://www.nist.gov/srd/nistsd4.htm

9. Maio, D., Maltoni, D., Cappelli, R., Wayman, J.L., Jain, A.K.:

FVC2002: Fingerprint veriﬁcation competition. In: Proceedings

of the International Conference on Pattern Recognition (ICPR),

pp. 744–747 (2002)

10. Ross, A., Dass, S., Jain, A.K.: A deformable model for ﬁngerprint

matching. Pattern Recognit 38(1), 95–103 (2005)

Individualization

▶ Fingerprint Matching, Manual

Influencing Factors

▶ Inﬂuential Factors to Performance

Influential Factors to Performance

KAO RU UCHIDA

NEC Corporation, Kawasaki, Japan

Synonym

Inﬂuencing factors

Definition

Factors that inﬂuence biometric performance. They

can be discussed from the following viewpoints: char-

acteristics of users (including the deﬁnition of im-

postors) and restrictions that come from practical

situations in which that biometric modality is used in

applications.

Introduction

When evaluating performance of biometric systems,

factors that inﬂuence performance (‘‘inﬂuencing fac-

tors’’) should be identiﬁed and analyzed because per-

formance is greatly affected by a w ide variety of

inﬂuencing factors. The same biometric device may

generate different test results if these inﬂuencing factors

differ. Controlling, recording, and reporting factors are

indispensable for executing repeatable performance

tests and predicting operational performance [1–3].

Influencing Factors

Identifying inﬂuencing factors: the following factors

should be considered at a minimum (see also Annex

Cof[1]):

1. Biometric sensor quality and characteristics

2. Biological or behavioral characteristics of the sub-

ject relevant to data collection (essential historical

or demographic data):

a. Invariable: Gender, ethnic origin

b. Variable:



Biological: age, body dimensions/anthropo-

metric data (height, weight, etc...), muscu-

loskeletal disorders

Influential Factors to Performance

751



Habitual/Social factors: smoking preference,

hairstyle, makeup, eyewear (glasses, con-

tacts, etc...), clothing



Occupation

3. Environmental factors applicable to the biometric

device, sensor, or application such as:

a. Temperature

b. Humidity

c. Illumination



Type (standard incandescent, ﬂuorescent,

tungsten halogen, reﬂector lamps, light

emitting diodes (LEDs), sunlight, etc...)

d. Noise

e. Position of sensor with regards to the user

4. Temporal change of the biometric features

5. Impact of

▶ forgery attempts on false acceptance,

particularly in behavioral modalities

6. Differences between the data capture and signal

processing subsystems used in the enrolment phase

and those used in the veriﬁcation/identiﬁcation

phase. (This text taken ISO/IEC TR 19795-3:2007

Information technology – Biometric performance

testing and reporting – Part 3: Modality-speciﬁc

testing is reproduced with the permission of the

International Organization for Standardization,

ISO. This standard can be obtained from any ISO

member and from the Web site of the ISO Central

Secretariat at the following address: www.iso.org.

In addition, the characteristics that affect performance

can be discussed from the following aspects:

1. The deﬁnition of impostors

2. Restrictions that come from practical situations in

which that biometric modality is used in applications

Characteristics of Impostors

There are two factors to consider the deﬁnition of

impostors:

1. Multiple biometric data from one person

2. Impostor attempts for behavior-based modalities,

such as voice or signature

For modalities in which multiple bio metric data can be

collected from one person, e.g., ﬁnger (10 ﬁngerprints

from one person) and iris (2 iris-images from one

person), a rule for permitting or prohibiting the use

of these data as impostor attempts should be clearly

deﬁned in performa nce testing.

In the case of behavior-based modalities, testing

results regarding impostor attempts (FMR or FAR)

may be inﬂuenced depending on whether (or how

much) an impostor tries to imitate an authorized

user’s behavior. For instance, the case in which an

impostor physically traces an authorized user’s signa-

ture that the impostor obtained differs signiﬁcantly in

FMR or FAR from the case, where the imposto r only

looks at the signature and imitates it. For these

modalities, a criterion regarding impostor attempts

should be deﬁned in performance testing.

Characteristics of Modality Specific to

Applications

In general, almost all modalities of biometrics are

used for user authentication, but some modalities are

expected to be used in different classes of applications,

for example, face-based identiﬁcation is widely used in

surveillance applications. While a user’s cooperation

can be expecte d in the former, it cannot be expected in

the latter case. Thus, variation of performance testing

should be considered depending on the way the mo-

dality is used in real applications.

These restrictions can be divided into two

classiﬁcations:

1. Factors relating to users, such as facial expressions

that affect the countenance of the face, wearing

eye-glasses or contact lenses for the iris, etc.

2. Factors relating to external environments that are

uncontrollable by the algorithm or system, such as

illumination change for face or background noise

for voice

These factors naturally affect the performance, and the

types and number of factors are different in each mo-

dality. These modality-dependent variations should be

considered in performance testing. In addition, a con-

cept of

▶ robustness test should be introduced to eval-

uate the sensitivity or robustness of the technology

toward environmental factors, in case the variation

of the factors strongly inﬂuences the observed

performance.

752

Influential Factors to Performance

Related Entries

▶ Performance Evaluation, Overview

▶ Performance Measures

▶ Performance Testing Methodology, Standardization

References

1. ISO/IEC 19795-1:2006 Information technology – Biometric per-

formance testing and reporting – Part 1: Principles and

framework

2. ISO/IEC 19795-2:2007 Information technology – Biometric per-

formance testing and reporting – Part 2: Testing methodologies

for technology and scenario evaluation

3. ISO/IEC TR 19795-3:2007 Information technology – Biometric

performance testing and reporting – Part 3: Modality-speciﬁc

testing

Information Content of Iris Images

▶ Iris Image Quality

Information Fusion

Information fusion is the merging of disparate informa-

tion, usually from different sources, which is combined

in order to achieve a wider perspective on a problem

than is possible if only one type of information were

considered. In the context of multimodal biometric

authentication, the term is used to describe how infor-

mation from different biometric modalities, for exam-

ple a facial image and a voice sample, can be combined

to make a better decision on whether the person seek-

ing to be authenticated is who he or she claims to be.

Depending on where in the system the information

from the two modalities is combined, one can distin-

guish between feature fusion, score fusion, and decision

fusion. Feature fusion is also known as early fusion,

while the latter two fusion types are know n collectively

as late fusion. Feature fusion is a method, by which the

features from the different modalities are fused into

combined feature vectors after the feature extraction

stage, but before the pattern recognition stage of the

authentication system. The pattern recognition process

then proceeds on the basis of the combined feature

vectors. Score fusion is a method, by which a separate

authentication system for each modality calculates a

mathematical likelihood or distance score between the

feature and the client model or template in that mo-

dality. The score fusion algorithm then combines the

two modality scores by means of a suitable formula,

such as the arithmetic mean of the two scores, and the

ﬁnal accept–reject decision is made on the basis of the

combined score. Decision fusion is a method, by which

each modality has its own independent authentication

process – from feature extrac tion throug h score calcu-

lation to ﬁnal accept–reject decision for that modality.

The overall accept–reject decision by the system is then

made through a simple logical combination of the two

modality decisions. Two decision fusion paradigms are

possible in a bimodal system: a strict system requires

an ‘‘accept’’ from both modalities for an overall accept

decision, while a generous system only requires an

‘‘accept’’ from either modality for an overall accept

decision. A system, which is to detect synchrony be-

tween the audio and video signals from a speaking face,

requires early fusion, while late fusion is sometimes

used when two off-the-shelf commercial authentica-

tion systems are combined and the user has no access

to the internal feature vectors of the two systems.

▶ Liveness Assurance in Face Authentication

Intelligent Agents

Software (sub-)systems that may act autonomously on

behalf of users of the system.

▶ Biometric Systems, Agent-Based

Interaction

The result an individual, using a biometric sensor,

creates by presenting biometric characteristics to a

Interaction

753

sensor. In terms of the general biometric model, this

occurs in the data capture silo. During an individual’s

interaction, the biometric sensor or device acquires an

image or signal of biometric characteristics of a person.

The goal of the presentation is for an individual to

present the biometric characteristics of high quality to

the sensor in a repeatable and consistent manner, so

that the subsequent signal processing sub-processes of

segmentation, feature extraction, and quality control

can successfully occur. See also attempt, presentation,

transaction, and general biometric model.

▶ Ergonomic Design for Biometric Systems

Interactive Voice Response (IVR)

Interactive voice response (IVR) is an automated, ‘‘self

service’’ telephony technology that interacts with a

caller to gather and dispense information, perform

transactions, or route the call to the appropriate recip-

ient. IVR systems can accept touchtone (DTMF),

spoken (speech recognition) input, or a combination

of the two modalities.

▶ Remote Authentication

▶ Speaker Recognition, Standardization

Interest Point, Region, Local Feature

In a way, the ideal local feature is a point as deﬁned in

geometry: having a location in space but no spatial

extent. In practice however, images are discrete with

the smallest spatial unit being a pixel and discretization

effects playing an important role. To localize features in

images, a local neighborhood of pixels need to be

analyzed, giving all local features some implicit spatial

extent. For some applications (e.g., camera calibration

or 3D reconstruction), this spatial extent is completely

ignored in further processing, and only the location

derived from the feature extraction process is used. In

those cases, one typically uses the term interest point.

However, in most applications those features also need

to be described, such that they can be identiﬁed and

matched, and this again calls for a local neighborhood

of pixels. Often, this neighborhood is taken equal to

the neighborhood used to localize the feature, but this

need not be the case. In this context, one typically uses

the term region instead of interest point. However,

beware: when a local neighborhood of pixels is used

to describe an interest point, the feature extraction

process has to determine not only the location of the

interest point, but also the size and possibly the shape

of this local neighborhood. Especially in case of geo-

metric deforma tions, this signiﬁcantly complicates the

process, as the size and shape have to be determined in

an invariant (covariant) way.

▶ Local Image Features

Interest Points

▶ Local Image Features

Interface

▶ User Interface, System Design

Intermediate Biometrics

▶ Biometric Sample Synthesis

Internal Identification

Identiﬁcation of a victim based on medical ﬁndings by

an autopsy, such as implan ts, and missing organs.

▶ Dental Biometrics

754

Interactive Voice Response (IVR)

International Association for

Identification

▶ Fingerprint, Forensic Evidence of

Interoperability

Interoperability in biometrics can be deﬁned as the

capability of a recognition system to operate with

data from different sources (e.g., data acquired using

different sens ors or features extracted using systems

from different vendors). Most biometric systems are

designed und er the assumption that the data to be

compared with are obtained from a unique source

and are restricted in their ability to match or compare

with biometric data originated from different sources.

As a result, changing the source may affect the perfor-

mance of the system.

▶ Fingerprint Databases and Evaluation

▶ Iris Device

Interoperable Performance

PATRICK GROTH ER

National Institute of Standards and Technology,

MD, USA

Definition

Accuracy of a biometric system that includes standar-

dized components from several suppliers.

In applications where components conform to

standardized interfaces and functional speciﬁcations,

it is possible to replace one component with another

from a different manufacturer. Although conformity

to speciﬁcations is a necessary condition for interoper-

ability, it is often not sufﬁcient, because the internal

algorithmic action of the component is usually not

regulated by the standard. Thus, a biometric detection

algorithm might underpe rform some others despite

being in conformity with the requirements. This arti-

cle suggests that the appropriate means of quantifying

biometric interoperability are to identify relevant per-

formance metrics, to measure them, and to certify

against them. For biometric sensors and also for detec-

tion, segmentation, and matching algorithms, these

metrics will be usually be failure-to-acquire and enroll

rates, and Type I and II recognition error rates.

Introduction

Biometric recognition is explicitly a two-phase opera-

tion: In veriﬁcation, a ﬁrst-encounter enrollment sam-

ple is compared with a second-encounter veriﬁcation

sample. Similarly in identiﬁcation, a new sample is

searched against a set of prior enrollments. If the sam-

ples are not captured and processed using the same

hardware and software, identically conﬁgured, the

issue of whether the various components are interop-

erable arises. While interoperability is a desirable and

necessary aspect of applications in which multiple

vendors sell equipment for capture, processing, and

matching, it rests on the availability of well-crafted

standards, and speciﬁcally, conformity to the various

components to those standards. Therefore, sensors

might have to conform to imaging speciﬁcations,

their outputs to image exchange standards, and their

transmission might require equipment implementing

standardized interfaces. The hazard in biometric appli-

cations is that a weak speciﬁcation or a lack of confor-

mity to a speciﬁcation might undermine the accuracy

of the whole recognition system.

Figure 1 depicts a general interoperable applica-

tions. It shows N different biometric capture devices

(BCDs) being used to acquire sample data that are then

converted from its raw captured biometric data block

(cBDB) format into a standardized biometric data block

(sBDB) format for enrollment. This is done by any of I

template generators. Later, these will be compared by

any of K comparison subsystems against veriﬁcation (or

identiﬁcation) records (sBDBs, in this case) produced

from any of J generators processing the output of M

BCDs.

This formalism is notional; it deﬁnes a ﬁve-

dimensional component space the last of which is

the comparison engine whose outputs suppor t mea-

surements of accuracy. Thus, any combination of ﬁve

different products can be tested. An interoperable

Interoperable Performance

755

component is then one that can be used in combina-

tion with others. Note that this deﬁnes biometric inter-

operability differently than in some other domains

where strict conformance guarantees performance.

For example, while a non conforming implementa-

tion of the PGP message standard [1] is likely to

give a deterministic and catastrophic failu re, a set of

ﬁngerprint minutiae automatically extracted from a

digital image of an analog trait (i.e., the ﬁnger) may

well give lower accuracy than those marked by a ﬁn-

gerprint examiner.

This article gives an overview of the biometric

interoperability problem and introduces the notion

that interoperability should properly be quantiﬁed

in terms of some relevant performance metrics. It

proceeds with examples of interoperability challenges,

which motivates the subsequent contribution on inter-

operability testing.

Interoperability Challenges

Successful recognition depends on t he interoperability

of all pieces of equipment used in generati ng both the

enrolled and recognition data records. This begins with

the acquisition process, and the primary requirement

is that the sensors are interoperable. This typically

means that the conversion of the analog human trait

into the digital sample produces a deﬁned or com-

monly understood representation of the original. The

following paragraphs give examples that undermine

interoperability of the three most common modalities.

Fingerprints: It is common in large-scale identity

management applications to acquire ﬂat impressions

of a subject’s ﬁngers, to associate those with a creden-

tial, and to verify against one or more of those ﬁngers.

While the ﬁngerprint data can be stored in conformity

to, for example, the ISO/IE C 19794-4 ﬁnger image or

19794-2 ﬁnger minutiae standards, subsequent veriﬁ-

cation attempts depend on the interoperability of the

capture devices with the original optical scanner. This

is one of the few areas of biometrics where sensors are

standardized: The U.S. Federal Bureau of Investigation

established a physical imaging speciﬁcation for optical

ﬁngerprint sensors. Known as Appendix F [2], this

document regulates the imaging capabilities of the

acquisition devices such that the representation of

the ﬁngerprint ridge structure is accurately represented

in the output image, and that the image is defensible in

criminal law enforcement. The speciﬁcation imposes

limits on parameters such as the optical resolution of

the device, the amount of geometric distortion, the

imaging area, and spatial uniformity.

Face: A face image collected at a di stance of 30cm

is unlikely to be interoperable with another acquired at

1m because of the presence of geometric ‘‘ﬁsh-eye’’

distortion. This would affect face recognition systems

whose internal representation of the face depends on

the relative spatial locations of the various anatomical

features. Although a nonlinear re-sampling of the

image could correct such distortion, the resulting spa-

tially varying resolution might undermine accuracy.

Another possible solution would be to formulate a

mathematical representation that is invariant to this

kind of distortion. The actual approach from the com-

mercial and user communities has been to regulate the

acquisition process via a formal technical standard.

This standard ISO/IEC 19794-5:2005 Face Image Data

requires distortion to be absent, and the amendment

ISO/IEC 19794-5/Amd. 1 Conditions for Taking Photo-

graphs for Face Image Data requires the subject to be

positioned at least 0.7m from the camera.

Interoperable Performance. Figure 1 Testing the performance of interoperable components.

756

Interoperable Performance

Iris: The interoperability of three iris cameras was

measured in the 2005 ITIRT trial [ 3]. The results of

cross-matching images using a single iris recognition

package showed that cross-camera accuracy was general-

ly worse than that for single-camera matching. Possible

causes for this, which was not asserted in the report,

might be the differences in the spectra of the infra-

red illuminants and in the compression applied post

capture.

Interoperable Data Formats

▶ Biometric data interchange standards have been

developed to advance interoperability of most of the

main biometric modalities. Standards exist for both

images and signals, and for ‘‘raw’’ sample data and

for processed data. The major extant internationally

standardized records are tabulated in Table 1. The

standards deﬁne a syntactic representation of the data

in question. These are usually compact binary encod-

ings of the data suitable for storage on a smartcard or

for transmission across a bandwidth-limited commu-

nications channel.

Any interoperability problems that could arise

from different implementations of the standards might

not be revealed until a test is run or a deployment occurs.

Although the possible problems are very speciﬁc to the

standards, the general case is that problems can be

expected when two very different sensors are used. For

example, in DNA typing, problems would occur if

the two sets of loci were disjoint. To avoid such effects,

the standards variously regulate the biometric acqui-

sition process.

Interoperability Testing

As various interoperability tests were staged around

the world [5, 3, 4], the Working Group 5 of ISO/IEC

JTC 1’s Subcommittee 37 on Biometrics, which stan-

dardizes biometric performance tests, started work on

interoperability testing. This culminated in 2007, in

ISO/IEC 19795-4 Interoperability Performance Testing

[6], which establishes procedures for the conduct of

tests such as those listed.

The standard requires a testing lab to establish

and identify one or more application speciﬁc ﬁgures

of merit, such as false non-match and false match

rates, and to report them in the manner presented in

Table 2. This shows the interoperability of INCITS

378 ﬁngerprint minutia template [7] generators and

matchers. It assumes an enrollment template generated

by equipment identiﬁed by the row label is later ver-

iﬁed against a template generated and matched by

equipment from the supplier identiﬁed in the

column headings. (Please note: The notable results

here are that the lowest false non-match error rates

are lowest when a single company executes all three

functions. That this ‘‘native mode’’ gives better perfor-

mance than the interoperable cases off the diagonal has

been attributed to idiosyncratic minutia placement

and selection strategies present in minutia detection

algorithms.)

The standard also establishes procedures for how

a certiﬁcatio n body could use mutually low error

rates in an interoperability test as a criterion to identify

a core group of interoperable products. The standard

then addresses how to maintain a certiﬁcation program

in which products are tested in regular ongoing testing

Interoperable Performance. Table 1 Biometric data interchange standards for various modalities

Standard Modality Processing

19794-2:2005 Fingerprint minutiae Template

19794-4:2005 Fingerprint Raw Image

19794-5:2005 Face Raw or Normalized 2D Image

19794-6:2005 Iris Raw or Polar Image

19794-7:2007 Signature time series Multivariate signal

19794-8:2006 Fingerprint skeleton Processed image

19794-9:2006 Vascular Raw Image

19794-10:2007 Hand geometry Binary silhouette Image

19794-13:2010 (est) Voice Raw Signal

19794-14:2010 (est) DNA Type Signal

Interoperable Performance

757

campaigns spanning perhaps several years, and in

which there might be systematic changes in the difﬁ-

culty of the test (due to environment, for example).

Sufficiency of a Biometric Data

Interchange Standard

Biometric data interchange standards support inter-

operability by allowing developers to implement pro-

ducts producing and processing records conforming to

a known format. Such standards deﬁne the syntactic

and semantic representations of the data. For example,

a depth value in a 3D face image might be encoded

as unsigned integer, but the precision might be com-

mercially important; if one supplier can accurately

determine depth to within 0.1 mm, then they would

lose accuracy if the standard provides only for depth

resolutions of 0.5 mm.

Together, the consensus speciﬁcations established

in a data interchange format standard might offer

less accuracy than a totally unconstrained repre-

sentation of the biometric data, and the quantiﬁ-

cation of such a loss goes to the sufﬁciency of a

biometric data interchange standard. A point to be

noted here is this terminology is that adopted in the

international biometric interoperability performance

testing standard [6] by answering the question, ’’does

standardized data offer accuracies approaching that of

unconstrained, nonstandard representations?’’.

Figure 2 depicts an ofﬂine testing methodology

for sufﬁciency. The MINEX I study [4] quantiﬁed

sufﬁciency for the (x, y, y , type) encoding of the min-

utiae deﬁned in INCITS 378 [7]. The excerpted results

Interoperable Performance. Table 2 Cross-vendor interoperability for a subset of the minutia detection and matching

algorithms evaluated in NIST’s MINEX [4] minutia interoperability baseline. The values are false non-match rates at a ﬁxed

false match rate of 0.01, for single ﬁnger veriﬁcation on a large ofﬂine database

Verification template and matcher provider

ABCD E G

Provider of enrollment template A 0.0136 0.0549 0.0458 0.0225 0.0641 0.0417

B 0.0218 0.0251 0.0385 0.0173 0.0402 0.0192

C 0.0357 0.0428 0.0225 0.0204 0.0519 0.0348

D 0.0207 0.0357 0.0301 0.0140 0.0485 0.0316

E 0.0236 0.0365 0.0340 0.0225 0.0301 0.0286

G 0.0300 0.0291 0.0447 0.0205 0.0390 0.0129

Interoperable Performance. Figure 2 Testing the sufficiceny of a data interchnage standard. The samples from a fixed

corpus are converted to both proprietary and standardized biometric data blocks (pBDBs and sBDBs, repsectively) and

then these are recognized by comparison subsystems from the same suppliers.

758

Interoperable Performance

of Table 3 show that proprietary implementations out-

perform the standard representation by less than the

variation between interoperable pairs observed in

Table 2.

Summary

Although interoperability can be supported by plac-

ing appropriate speciﬁcations on the various com-

ponents, particularly sensors, biometric recognition

performance tests are sometimes the only means of

quantifying interoperability. More importantly, accu-

racy testing is the most operationally relevant measure

of interoperability.

Related Entries

▶ Interoperability

▶ Performance Testing

▶ Standards

References

1. et al.: RFC 4880 - OpenPGP Message Format (2007).

2. Hopper T., et al.: IAFIS Image Quality Speciﬁcations, EBTS

Appendix F. Tech. rep., FFBI Criminal Justice Information

Serv ices Divis ion (2008). Http://www.fbibiospecs.org

3. Thieme M., et al.: Independent testing of iris recognition tech-

nology ﬁnal report. Tech. rep., International Biometric Group

(2005). Http://www.biometricgroup.com/ITIRT/

4. Grother P.J., et al.: Minutiae Exchange Interoperability

Test MINEX - Performance and Interoperability of

the INCITS 378 Fingerprint Template. Tech. Rep. NIST

7296, National Institute of Standards and Technology,

Gaithersburg, Maryland (2006). Available at http://ﬁngerprint.

nist.gov/minex

5. Campbell J., et al.: Seafarers’ Identity Documents - Biometric

Testing Campaign Report. Tech. rep., International Labour

Organization (2005)

6. JTC 1, SC37 Biometrics, Working Group 5: ISO/IEC 19795-4

Interoperability performance testing (2007). http://webstore.

ansi.org

7. INCITS M1, Biometrics: INCITS 378:2004 Fingerprint minutia

format, 1 edn. (2004). URL http://webstore.ansi.org. American

National Standard for Information Technology

Intraclass

Intraclass refers to different instances of the same sub-

ject seen under different viewing conditions, illumina -

tions, etc. Ideally, the extracted features should be

similar for different instances of the same subject.

▶ On-Card Matching

Intricated

Intricated is having many complexly arranged ele-

ments. A secret key and ﬁngerprint minutia are intri-

cated into a template so that it is not possible to get the

key or the minutiae from the tem plate.

▶ Fingerprints Hashing

Intricated Biometrics

▶ Fingerprints Hashing

Interoperable Performance. Table 3 False non-match rates at fixed rate of 0.01 for fingerprint verification algorithms

using fully proprietary and formally standardized templates. The proprietary accuracies would only be available in an

interoperable application if the images were exchanged

Kind of template

Provider of template generator + matching algorithm

ABD EG

Proprietary 0.0089 0.0189 0.0089 0.0251 0.0047

Standard 0.0136 0.0251 0.0140 0.0301 0.0129

Intricated Biometrics

759

Callas J.,

Http://tools.ietf.org/html/rfc4880