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

Подождите немного. Документ загружается.

Comparison

▶ Palmprint Matching

Comparison Prints

These known exemplars of ﬁngerprints are taken from

donors under controlled conditions. The most com-

mon method used for recording these prints is ink on

paper; however, this method is gradually being

replaced by live scan technology. For law enforcement

purposes, the prints of all 10 ﬁngers are commonly

recorded. For other purposes such as border control

or passports, this may be limited to one or four ﬁngers.

▶ Fingerprint Matching, Manual

Complementary Metal Oxide

Semiconductor (CMOS)

CMOS is a widely used architecture for integrated

circuits, particularly semiconductor memory circuits.

CMOS has been adapted for use in image sensors

and is a competitor to CCD image sensors. This is a

newer sensor technology; its compatibility with the

large scale integration techniques developed for semi-

conductor memory is a powerful advantage. CMOS

and CCD technologies are battling for dominance in

the marketplace.

▶ Iris Device

Compliance

▶ Conformance Testing for Biometric Data Inter-

change Formats, Standardization of

Computational Iris Recognition

Systems

▶ Wavefront Coded

Iris Biometric Systems

Concept Drift

The target concept in a machine-learning task might

change over time in terms of distribution, description,

properties, etc. Often the changes are unpredictable,

which brings problems to learning systems without

self-adaptive mechanisms because the predictions

of such systems might become less accurate as the

time passes.

▶ Incremental Learning

Confidence Interval

A 100(1  a)% conﬁdence interval for some parame-

ter y is a range of values (L, U) such that P(yE(L, U)) =

1  a, where L and U are random variables.

▶ Test Sample and Size

Configural Processing

Conﬁgural processing is a speciﬁc mechanism that may

evolve over time in which features are considered in

relation or context with other features or areas. Con-

ﬁgural processing typically allows more information to

be extracted from each location or feature.

▶ Latent Fingerprint Experts

190

Comparison

Configuration Issues, System Design

KAI CAO,JIE TIAN,YANGYANG ZHANG,XIN YANG

Institute of Automation, Chinese Academy of Sciences,

Beijing, People’s Republic of China, China

Synonym

Setting

Definition

Conﬁguration is to select appropriate parameters

and/or fusion rules for an individual user or globally

for all users of a system under a speciﬁc environmental

condition (including the environment, the device, the

received data, and the biometric algorithm).

Introduction

A ty pical biometric system can be divided into two

distinct stages [1]: the enrollment stage which includes

capture, feature extraction and template generation,

and the veriﬁcation or identiﬁcation stage which

comprises capture, feature extraction, matching and

decision, as illustrated in Fig. 1. In the enrollment

stage, for each user, a biometric trait is captured and

processed to present it with a feature set. Then a

template is generated and labeled with the user’s iden-

tity. In the veriﬁcation or identiﬁcation stage the input

biometric trait is processed as above and the system

outputs the matching score or decision. Each compo-

nent of the system in the two stages is cri tical and may

be affected by the environmental condition. Care must

be taken in any speciﬁc application since different

kinds of applications focus on different requirements.

Due to the variety of biometric system application in

various environments and situations, conﬁguration of

selecting appropriate parameters and/or fusion rules to

the biometric system is essential in order to satisfy

different requirements.

For the component of capture, each device will

have certain criteria to conﬁgure the capture process.

For example, in a ﬁngerprint device, the qualit y of the

captured ﬁngerprint image for enrollment should be

high enough to ensure the reliable characteristic fea-

tures (e.g., minutia) of the ﬁngerprint. For face recog-

nition devices, the person is usually required to be in a

standard position directly facing the capture device.

Threshold levels are also used in template generation

process, where they determine the similarity required

to make samples to be able to generate repeatable

Conﬁguration Issues, System Design. Figure 1 A simplified verification system.

Configuration Issues, System Design

191

templates. All of these are unique considerations for

biometric system.

Another important component is the decision pro-

cess which mainly controls the security and accuracy.

Security and accuracy is mainly controlled by the deci-

sion process. For a single biometric veriﬁer the only

parameter need to be determined is the threshold

which can be conﬁgurable by an administrator or it

may be ﬁxed by the biometric system either for an

individual user or globally for all users. If the matching

score produced by the biometric system is larger than

the threshold then the user is accepted, otherwise the

user is rejected. However, the performa nce of the bio-

metric system may be affected by the scenario of the

application, environmental conditions, amount of user

population, and many other factors. For example, iris

recognition may depend on light levels, voice recogni-

tion may depend on ambient sound levels, and atmo-

spheric dust levels may affect ﬁngerprint devices. On

the other hand, recognition based on any one of these

modalities may not be sufﬁciently robust or else may

not be acceptable to a particular user group or in a

special situation because of restricted degrees of free-

dom and unacceptable error rates. One way to improve

the performance is to install multiple sensors that

capture different biometric traits, and this approach

either focuses on a ﬁxed set of traits or seeks greater

ﬂexibility through the implementation of systems

which are more generically adaptable and reconﬁgur-

able. Conﬁguration of selecting appropriate thresholds

and fusion rules for this type of system increases com-

plexity and raise questions, and is therefore regarded as

a major challenge.

Configuration in Capture Process

Due to the changes in the environmental conditions

(e.g., light level, dust, humidity, and cleanliness of the

biometric capture device), the biometric system may

capture biometric data of poor quality, which, especially

in the enrollment stage, is responsible for many of the

most matching errors in biometric systems so that it will

limit the accuracy of the biometric system. For example,

Tan e t al. [2] pointed out, from the experiments on the

ORL face database that the performance of an eigen-

face-based face recognizer drops quickly when the en-

rolled templates become poorly representative. The

problem, however, may be alleviated by the control of

the quality of the biometric data. If the input biometric

data quality measurement is lower than a threshold

conﬁgured by the system or the administrator,

the device will reject the biometric data of poor

quality and the capture process may be repeated.

Quality level of the captured biometric data is expected

to be high during the enrollment stage, since it

forms the basis against which all further biometric

matchings are made. Therefore, the quality threshold

should be higher in the enrollment stage than

that in the veriﬁcation or identiﬁcation stage. Some

other criteria should also be considered. For example,

if the area of captured ﬁngerprint is too small it will

impact false accept rate (FAR) and false reject

rate (FRR).

From the consideration of security, the biometric

system may be attacked by fake or artiﬁcial biometric

traits. This case exists especially in ﬁngerprint captur-

ing. Fingerprint capture devices are deceived probably

by well-duplicated fake ﬁngers [3]. Therefore, it is

necessary to detect fake bio metric traits in order to

ensure that only live biometric traits are capabl e of

generating templates for enrollment and recognition.

The biometric system may be conﬁgured to select

approaches and parameters to ensure that the captured

sample comes from a live human being.

Single-modal biometric system has limitations in

terms of enrollment rates, and susceptibility to spoof-

ing. A recent report [4] by the National Institute of

Standards and Technology (NIST) to the United States

Congress concluded that approximately two percent

of the population does not have a legible ﬁngerprint

and therefore cannot be enrolled into a ﬁnger-

print biometrics system. However, this problem can

be solved by employing multibiometrics in a layered

approach. Therefore it is essential for the bio-

metric system to provide an option to select a different

biometric trait or employ a combination of these bio-

metric traits to make the system adapt to different

scenarios.

Configuration in Template

Generation

As pointed out clearly by Uludag et al. [5], in real

operational scenarios, input biometric data can exhibit

192

Configuration Issues, System Design

substantial variations compared to the templates col-

lected during the enrollment stage of users. In other

words, there may be a large intra-class variability, due

to changes in the environmental conditions (e.g., illu-

mination changes), aging of the biometric traits, varia-

tions of the interaction between the sensor and the

individual (e.g., poor ﬁnger placement and facial ex-

pression of a person’s face), all of which will limit

the performance of the biometric system. In order to

obtain good performance, conﬁguration may be re-

quired to preserve distinct and repeatable biometric

features from the user. This process is critical from a

security evaluation point of view, since the level of

uniqueness inherent in a template will inﬂuence FAR

and FRR of the system. The ﬁrst conﬁguration parame-

ter is the number of the templates, since the system

accuracy can be improved by collecting multiple tem-

plates with different variation (e.g., different pose of a

person’s face) in the enrollment stage but the storage

requirement will be huge if the number of the template

becomes very big.

There are basically three possibilities of where to

store the template: on a token, local unit, or on a central

server. Each of these locations has its own advantages

and disadvantages: (1) To store the template on a por-

table token such as smart card does not need to traverse

the network. The user carries the template from loca-

tion to location. The compare process will be very fast

because only the template on the card needs to be

considered. One drawback with this storage is that

template on the smart card can be read by unauthorized

individuals and the template will be stolen. (2) Storing

on a central sever overcomes the problem of users

authenticating from multiple locat ions. The input

data will have to be transferred through the network.

Therefore it provides an opportunity for a third

party to intercept the data transfer and duplicate the

biometrics data. The input data may be compared with

every record data stored to identify the user. The speed

will be very slow as the number of users grow. (3) The

computation ability of local unit seems to be middle

ground between central sever and small card. And dis-

tributively storing the data prevent a focal point of

attack for malicious hackers. However, with this storage

the security may be lacking because the template

could be found on the hard drive.

On the other hand, performance of biometric rec-

ognition systems may degrade quickly when the time

interval between the input biometric trait and the ref-

erence template is long because biometric features can

change over time. The biometric system may be con-

ﬁgured to update the reference template by re-enroll or

during the matching operations, improve the refer-

ence template by merging and averaging minutiae of

multiple biometric traits.

Configuration in Decision Process

Conﬁguration in decision process for a biometric system

can be classiﬁed coarsely into tw o categories in terms of

biometric traits adopted by the system: single-modal

biometric conﬁguration and multi-modal biometric

conﬁguration. Both of these two classes of conﬁguration

are mainly to set the thresholds for each biometric device

and/or the fusion rules algorithm among the sensors.

The ability to estimate veriﬁer error rates is very impor-

tant for such conﬁguration. Without knowledge of how

well the system works under current conﬁguration para-

meters, there is no wa y of knowing whether parameters

should be changed and which parameter should be

adjusted. Estimating veriﬁer error rates provides feed-

back to remedy this situation.

Error Rate Estimation

There are mainly six important error rates [6]usedto

measure a biometric system: failure to acquire rate, failure

to enrol l rate, false match rate (FMR), false non-match

rate (FNMR), false acc ept rate (F AR), and false reject rate

(FRR). FAR and FRR are two of the most important

indications of error rates of a biometric system. FAR is

the expected proportion of transactions that a transaction

will be erroneously accepted by the biometric system

when it should have been rejected. FRR is the expected

proportion of transactions that a transaction will be

erroneously rejected when it should have been accepted.

From a security aspect, FAR has much more rele-

vance than FRR. However, FRR can be considered as a

measure of inconvenience but also a measure of avail-

ability, and needs to be kept within acceptable limits

for the intended application. So, knowledge of FAR

and FRR can be used to select appropriate conﬁgura-

tion parameters. FRR is generally straightforward to

determine, while FAR is often difﬁcult to analyze in

Configuration Issues, System Design

193

operational system. FRR at the chosen decision thresh-

old can be estimated by asking authentic users to

report when they are rejected. This assumes that each

time when an authentic user is rejected, he or she will

report only once. Another way to estimate FRR is to

represent transaction of authentic users to the system

and record the probabilities of failure (or matching

scores). The simplest way to estimate FAR and FRR

simultaneously is to use stored transactions. Each

transaction is compared with the other transaction of

the same user and the probability of failure (or match-

ing score) is record. All of the records form the genuine

distribution. Similarly, the impostor distribution can

be computed by comparing each transaction with

transactions of other people using the stored transac-

tions and recording the probabilities of failure (or

matching scores). Then the obtained probability den-

sity (or matching score) curve can be used to estimate

FRR and FAR at all threshold settings. All of these

estimations should be done unde r the concrete envi-

ronment, since biometric traits and performance of the

system may be affected by their physical environmental

condition. For example, a device to read iris patterns

relies obviously on the ambient lighting conditions.

This means that the physical environment needs to be

deﬁned as part of the biometric system’s conﬁguration,

and also that tests should be conduct under the same

or similar environmental condition.

Because of the expense of collecting large databases,

performance of the devices or algorithms is usually

measured on relative small databases. Wayman et al.

[7] speciﬁes four methods to evaluate FAR and FRR

for large-scale identiﬁcation system: (1) extrapolation

from experiences; (2) identiﬁcation as a succession of N

veriﬁcation; (3) extrapolation with extreme value; (4)

extrapolation when the distance can be modeled. In fact,

different environmental conditions such as illumination

and humidity can also result in variation of FAR and

FRR. Beattie et al. [8] proposed a structured approach

based on expectation-maximization (EM) algorithm

for estimating error rates.

Single-Modal Biometric Configuration

In theory, matching scores of authentic users should

always be higher than that of impostors. If this would

be true, a single threshold that separates the two groups

of scores could be used to differentiate between

authentic users and impostors. Due to several reasons,

this assumption isn ’t true for real-world biometric

systems. In some cases, impostor patterns generate

scores that are higher than the scores of some authentic

patterns. For that reason, it is a fact that there is a

threshold that needs to be determined to control the

security and convenience of the biometric system. If the

threshold is too high, the veriﬁer will mostly reject

decision regardless of whether the claimant is genuine

or an impostor, the system is very secure but inconve-

nient. On the other hand, if the threshold is too low, it is

convenient but insecure. Choosing a suitable threshold

for a particular environment can be completed by ana-

lyzing the error rates on the training database captured

in the same environment. The matching scores pro-

duced by the biometric system under the training data-

base give a distribution for authentic users and another

for impostors. When a threshold is set, FAR and FRR

are simultaneously ﬁxed, as illustrated in Fig. 2.

FAR and FRR are dependant on the adjustable

threshold. If the value of threshold is increased,

the proportion FAR will increase, while FRR will de-

crease. Other wise, FAR will decrease and FRR will

increase. This relationship between the two error rates

is characterized by a receiver operating characteristic

(ROC) curve in which FAR is plotted against FRR by

varying the threshold, as shown in Fig. 3. Trade-off

between FAR and FRR is often achieved by selecting

an appropriate threshold so that the two rates can

satisfy both the prescribed security and convenience.

The biometric system (e.g., physical access control)

may be distributed over multiple locations where the

environmental conditions may differ from each other.

For this type of system, each biometric veriﬁer must be

conﬁgured with an appropriate threshold according to

the performance under the environmental condition.

In order to increase the reliability of single trait, the

combination of different biometric matchers can be

performed at the score level.

Combination of multiple matching algorithm is

performed typically at the score level, and different

fusion techniques (e.g., average, product, sum, max

etc) have been applied successfully [7, 8]. The aim of

multiple biometrics combined at the score level is to

produce new scores whose distributions for genuine and

impostor users exhibit a higher degree of separation

than those produced by individual matchers. Thus, by

varying the decision threshold, a better trade-off be-

tween FAR and FRR can be attained.

194

Configuration Issues, System Design

Multi-Modal Biometric Configuration

In order to make the overlap of the distribution graphs

of FAR and FRR in multi-modal biometric sy stem, as

little as possible, scores fusion rules can be adopted

according to the conﬁdentiality capture environment

and recognition success rate of individual matchers.

For this kind of multimodal biometric system not only

the thresholds, but also the fusion rules need to deter-

mined. Knowledge of the characteristics of each bio-

metric trait and sensor will be helpful to design an

effective system. For example, from the experiment of

Conﬁguration Issues, System Design. Figure 2 FAR and FRR on normalized distributions graphs associated with a

specific threshold.

Conﬁguration Issues, System Design. Figure 3 Receiver Operating Characteristic (ROC) curve.

Configuration Issues, System Design

195

Ross et al. [9], the ﬁngerprint and face biometrics

generated a relatively small enroll failure rate, while

the performance of the voice is much more stable than

the other two modalities once a satisfactory enrollment

has been achieved.

The most commonly adopted approach in multi-

modal biometric system is to use the same fusion rule

and the same decision threshold for all users [9], the

main idea of this system is to treat all matching scores

from genuine users as one single class while all match-

ing scores from imposter users as the other one. Ross

et al. [9] combine the matching scores of three traits

(face, ﬁngerprint, and hand geometry) to enhance the

performance of a biometric system. Experiments indi-

cate that weighted sum rule outperforms other three

techniques (sum rule, decision tree, and linear discrim-

inate analysis) in terms of ROC curves. However, they

do not mention how to conﬁgure the threshold. M. C.

FairHurst [10] described an approach that used genetic

algorithm (GA) to select appropriate parameters in-

cluding weights and threshold to evolve efﬁcient con-

ﬁguration using the Total Error Rate (FAR + FRR) of

the overall system as an evaluation criterion.

Relatively, another new approach is using multiple

fusion rules (each individual user correspond to a fusion

rule or/and multiple decision thresholds (each indi-

vidual user correspond to a threshold). Anil K. Jain

[11] proposed a user-speciﬁc multimodal biometric

system in which the common threshold is computed

using the cumulative histogram of impostor matching

score corresponding to each user and the user-speciﬁc

weights associated with each biometric are selected

by minimizing the total veriﬁcation error. Toh et al.

[12] improves this method using multivariate polyno-

mial fusion model for each user. In other words, the

system conﬁgures a personalized decision hyperplane

(including thresholds) for each use r.

Summary

Conﬁguration plays an impor tant role in the biometric

system. A threshold in the decision process controls

the trade-off between the security and the convenience.

A multi-modal biometric system can utilize the pre-

dominance of each biometric trait and allow a more

reliable biometric system. Accurate error estimation

information would be useful to conﬁgure appropriate

thresholds and/or fusion rules which w ill make the

system more effective. Appropriate conﬁguration will

make the biometric system more robust, adaptive, and

effective.

Related Entries

▶ Fusion, Score-level

▶ Multiple Classiﬁer Systems

▶ Multi-modal sy stems

▶ Multibiometrics

▶ Performance Evaluation, Overview

References

1. Ross, A.A., Nandakumar, K., Jain, A.K.: Handbook of Multi-

biometrics. Springer, New York (2005)

2. Tan, X., Chen, S., Zhou, Z.-H., Zhang, F.: Face recognition from

a single image per person: a survey. Pattern Recognit. 39(9),

1725–1745 (2006)

3. Matsumoto, T.H., Yamada, K., Hoshino, S.: Impact of artiﬁcial

‘gummy’ ﬁngers on ﬁngerprint systems. In: Proceedings of SPIE,

San Jose, CA, vol. 4677 (2002)

4. NIST report to the United State Congress. Summary of NIST

standards for biometric accuracy, tamper resistance, and

interoperability. http://www.itl.nist.gov/iad/894.03/NISTAPP_

Nov02.pdf (2000). Accessed 13 Nov 2000

5 Umut, U., Ross, A., Jain, A.K.: Biometric template selection and

update: a case study in ﬁngerprints. Pattern Recognit. 37(7),

1533–1542 (2004)

6. Common Criteria Biometric Evaluation Methodology Working

Group - United Kingdom: Common Criteria - Common Meth-

odology for Information Technology Security Evaluation - Bio-

metric Evaluation Methodology Supplement. http://www.cesg.

gov.uk/policy_technologies/biometrics/media/bem_10.pdf. Ver-

sion 1.0, Aug., (2002)

7. Wayman, J., Jain, A., Maltoni, D., Maio, D.: Biometric Systems:

Technology, Design and Performance Evaluation. Springer,

New York (2005)

8. Beattie, M., Vijaya Kumar, B.V.K., Lucey, S., Tonguz, O.K.: Au-

tomatic conﬁguration for a biometrics-based physical access

control system. In: International Workshop on Biometric Rec-

ognition Systems (IWBRS), Beijing, 22–23 Oct, pp. 241–248

(2005)

9. Ross, A., Jain, A.K.: Information fusion in biometrics. In:

Proceedings of International Conference on Audio- and Video-

Based Biometric Person Authentication (AVBPA), Sweden,

pp. 354–359 (2001)

10. Fairhurst, M.C., Deravi, F., George, J.: Towards optimised imple-

mentations of multimodal biometric conﬁgurations. In: IEEE

International Conference on Computational Intelligence for

Homeland Security and Personal Safety, Venice, Italy, 21–22

July 2004

196

Configuration Issues, System Design

11. Jain, A.K., Ross, A.: Learning user-speciﬁc parameters in multi-

biometric system. In: Proceedings of International Confer-

ence on Image Processing (ICIP), Rochester, New York, USA

pp. 57–60 (2002)

12. Toh, K.-Ann, Yau W.-Y.: Some learning issues in user-speciﬁc

multimodal biometrics. In: Eighth International Conference on

Control, Automation, Robotics and Vision Conference (ICARCV),

Kunming, China, 6–9 Dec, vol. 2, pp. 1268–1273 (2004)

Conformance Testing

Conformance testing is the process of capturing the

technical description of a speciﬁcation and measuring

whether an implementation faithfully implements the

speciﬁcation by achieving conformance to the techni-

cal description of the speciﬁcation. Conformance is

deﬁned generally as the fulﬁllment by a product, pro-

cess, or service of all relevant speciﬁed requirements.

▶ Conformance Testing for Biometric Data Inter-

change Formats, Standardization of

Conformance Testing for Biometric

Data Interchange Formats,

Standardization of

JOHN W. M. CAMPB ELL

,GREGO RY ZEKTSER

Bion Biometric s Inc., ON, Canada

US Depar tment of Defense, Biometrics Task Force

Synonyms

Conformity; Compliance

Definition

The development of standardized methods and proce-

dures for

▶ conformance testing of products or sys-

tems that claim to satisfy the requirements of one or

more of the standardized biometric data interchange

formats.

Concepts in Conformance Testing

A national or international standard consists of a set

of requirements and frequently a set of recommenda-

tions. The requirements speciﬁed in the standard are

traditionally classiﬁed in three categories: mandatory

requirements to be observed in all cases, optional

requirements to be observed if so chosen by an imple-

menter, and conditional requirements to be observed

under speciﬁc conditions. A product, process, or system

that fully satisﬁes the requirements of the standard is

described as being conformant to that standard. Con-

formance testing is the method that is used to determine

if the product, process, or system satisﬁes the require-

ments. The precise nature of these requirements differs

substantially from one standard to another, but in all

cases, there are certain important concepts.

The product, process, or system being tested is

known as an

▶ implementation under test or IUT. It

does not need to satisfy every requirement and recom-

mendation of a standard, only those that that are de-

ﬁned as mandatory. In certain cases, the mandatory

requirements may be different for different applications

or purposes. In the case of conformance testing for

biometric data interchange format standards, for exam-

ple, an IUT may be designed to produce biometric data

interchange records or to use biometric data inter-

change records or both. In each case, the requirements

that are tested for conformance would be different.

No conformance test can be complete or perfect.

Ultimately, it is only possible to prove that an imple-

mentation under test is nonconformant. The goal of

conformance testing is therefore to capture enough of

the requirements of the base standard and test them

under enough conditions, that any IUT that passes the

conformance test is likely to be conformant. Frequently,

there are inherent problems with the underlying stan-

dards that only become apparent during conformance

testing. For instance some areas may be undeﬁned

(so that the speciﬁcation of these areas is left to each

vendor) or ill-deﬁned (so that there is a contradiction

between parts of the base standard or an easy misinter-

pretation caused by the wording of the base standard).

The latter problem may be resolved by an amendment

to or revision of the standard, but the former problem

may be difﬁcult to resolve.

Conformance testing does not guarantee interoper-

ability; it can only provide a higher level of conﬁdence

that interoperability can be achieved. Although the

Conformance Testing for Biometric Data Interchange Formats, Standardization of

197

ultimate goal of standards is to allow different products

and systems to work together, even two products that

are conformant to the same standard may have difﬁcul-

ty working together. This is because it is usually impos-

sible for a standard to specify every aspect of the

operation of a product or system. On the other hand,

unless systems are conformant to a well written stan-

dard, then it is very unlikely that they will be interoper-

able. Thus conformance testing is a critical element in

assuring interoperability, even if it is not the only one.

Motivation for the Development of

Conformance Testing Methodology

Standards

As increasing numbers of biometric standards have

been developed in recent years, more and more pro-

ducts have become available that claim to be confor-

mant to the standards. This is par ticularly true in the

area of standardized biometric data interchange for-

mats which are standard methods of encoding biomet-

ric data for various technologies, incl uding 2D face, 3D

face, ﬁnger image, ﬁnge r pattern, ﬁnger minutiae, iris

image, vein pattern, signature/sign, hand geometry,

etc. Theoretically, those products that support the rel-

evant standard for a given technology should be able to

work together, so that an end user of biometrics can

mix products from a variety of vendors or suppo rt

interoperability among different systems.

Although vendors provide products and systems in

good faith believing that they conform to a standard, if

there is no corresponding conformance testing meth-

odology standard, then there is no clear method for

them to be able to verify this. Similarly, end users of

biometric products cannot know with conﬁdence if the

products and systems they are using actually conform

to the standards unless a formal conformance testing

methodolog y standard exists and can be used to per-

form conformance testing on those products and sys-

tems in a reliable and repeatable manner.

Elements Required in Conformance

Testing Methodology Standards for Data

Interchange Formats

In order to formally describe conformance testing for

data interchange formats, it is necessary to identify a

language to deﬁne the context of conformance

testing and conformance claims. Therefore a number

of specialized terms have been developed. Many of them

relate to the fact that there are lots of different types of

testing that can be deﬁned for different levels and types

of conformance. In the standardization process that has

taken place in the US at INCITS M1 – Biometrics and

internationally at ISO/IEC JTC 1/SC 37 – Biometrics,

the following key elements have been deﬁned.

Test Assertion – The speciﬁcation for testing a

conformance requirement in an IUT in the for ms

deﬁned in a conformance testing methodology stan-

dard. Test assertions are short speciﬁc statements that

encapsulate a single requirement for a particular stan-

dard. A harmonized assertion description language has

been developed for data interchange format confor-

mance testing so that the assertions can be expressed

in a simp le form, regardless of the speciﬁc data inter-

change format being addressed.

Level 1 Testing – A set of test methods within the

conformance testing methodology that validates ﬁeld by

ﬁeld and byte by byte conformance against the speciﬁ-

cation of the

▶ Biometric Data Interchange Record as

speciﬁed in the base standard, both in terms of ﬁelds

included and the ranges of the values in those ﬁelds.

Level 2 Testing – A set of test methods within the

conformance testing methodology that tests the inter-

nal consistency of the Biometric Dat a Interchange Re-

cord (BDIR) under test, relating values from one part

or ﬁeld of the BDIR to values from other parts or ﬁelds

of the BDIR.

Level 3 Testing – A set of test methods within the

conformance testing methodology that tests that a

Biometric Data Interchange Record produced by an

IUT is a faithful reproduction of the input biometric

data that was provided to the IUT.

Type A – Produce Conformant BDIR (Type A or

PCB) – A conformance claim by an IUT that it is a

conformant BDIR, or can create conformant BDIRs

from an appropriate input data.

Type B – Use Conformant BDIR (Type B or UCB) –

A conformance claim by an IUT that it can read con-

formant BDIRs, interpret them correctly, and perform

its desired function upon them.

Issues Related to Testing Levels

It is obvious from the carefully deﬁned terminology

listed above that there are issues that have led the

198

Conformance Testing for Biometric Data Interchange Formats, Standardization of

standardization bodies to separate the different levels

and types of testing. The main consideration is the

need for a balance between the importance of deliver-

ing conformance testing methodology standards

that are meaningful and that can be used to support

testing and the desire to thoroughly test all aspects of

each data interchange format standard.

The ﬁrst issue is the fact that data interchange

format standards are mostly focused on the structure

and content of the BDIR. This means that the test

assertions for Level 1 testin g can be simply developed

by analyzing the explicit requirements of the stan-

dard. Test assertions for Level 2 testing may require

consideration of the implicit requirements of the

standard, but they can still be deﬁned quite speciﬁ-

cally. S ome expert s prefer to stat e that Level 1 testing

supports the syntactic requirements of the standard

and Level 2 testing supports the semantic require-

ments of the standard. Unfortunately, some semantic

requirements can only be addressed through Level 3

testing, and because of the inherently uncertain na-

ture of biometric data, it is very difﬁcult to establish a

standardized method of determining whether a BDIR

is or is not a faithf ul reproduction of the input bio-

metric data used to produce it. Human b iometric

characteristics vary with every presentation to a bio-

metric system and there is debate among experts on

exactly how to deﬁne the relationship between the

BDIR and the input characteristic, especially when it

comes to acceptable levels of accuracy in the repre-

sentation. For this reason, Level 3 testing is still an

area of research and has not been included in the

conformance testing standards that are currently pub-

lished or under development.

The second issue relates to the fact that the BDIR

itself is the focus of the biometric data interchange

format standards. It is therefore easy to test claims of

Type A conformance, since the output BDIRs can be

tested at least for Level 1 and Level 2 conformance. An

IUT that claims Type B conformance, however, needs to

interpret the BDIRs correctly and perform its appropri-

ate function upon them. Since this function may be to

use them for biometric matching, to display them for

human examination, to convert them to another format

or potentially a whole host of other things, it is very

difﬁcult to determine how best to test such claims of

conformance. One option is to force IUTs to also sup-

port speciﬁc functions of usage that would only be used

in Type B conformance testing, but so far this idea has

not been popular among biometric vendors or standar-

dization experts. It remains to be seen how Type B

conformance testing will be addressed in the future.

Conformance Testing

Standardization – Current State

The need for standardized and commonly accepted

conformance testing methodologies for Biometric data

interchange formats has been recognized by the Nation-

al and International Standards Bodies on Biometrics. In

February 2005, INCITS M1 initiated the development of

a multi-part American National Standard on confor-

mance testing methodology for Biometric Data Inter-

change Formats. This project is based on an extensive

analysis of the data format requirements speciﬁed in the

base data interchange format standards, and is

structured to take advantage of the commonalities

found in the testable requirements as well as in the

conformance test methods and procedures. The result-

ing structure of this multi-part standard is as follows:

1. Part 1: Generalized Conformance Testing

Methodology

2. Part N: Modality-speciﬁc Testing Methodology

(e.g., Part 2: Conformance Testing Methodology

for Finger Minutiae Data Interchange Format)

The Generalized Conformance Testing Methodology

contains the elements of the testing methodology that

are common to all the data interchange formats (i.e.,

those elements that are modality independent). These

elements include deﬁnitions of terms, descriptions of

levels and types of testing, general requirements of test

reports, speciﬁcation of the assertion deﬁnition lan-

guage, general test procedures, etc.

Each individual Part contains elements of the test-

ing methodology speciﬁc to its respective modality.

These elements include speciﬁc deﬁnitions of terms,

speciﬁcations of test assertions, test criteria, modality-

speciﬁc elements of test reports, test procedures, etc.

At the time of preparation of this paper, the Part 1

of this multi-part standard has been published as

INCITS 423.1, and other parts are in various stages of

development ranging from publication stage to Public

Review stage. It is expected that most of the Parts of

this standard will be publicly available in 2008.

While the development of the INCITS 423

was underway, Working Group 3 of ISO/IEC JTC 1/SC

Conformance Testing for Biometric Data Interchange Formats, Standardization of

199