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

Подождите немного. Документ загружается.

be pre-aligned by the modeled orientation ﬁeld

(global feature). Then the similarity score is calcu-

lated based on the minutiae (local feature). Pre-

alignment is more efﬁcient while matching on the

large-scale database.

Performance Evaluation

Performance evaluation is necessary for understanding

the limitations and advantages of a ﬁngerprint match-

ing algorithm and addressing its appropriate applica-

tions. The performance can be evaluated from different

aspects: accuracy, resource consumption, scalability

and sensor interoperability.



The accuracy of matching is evaluated based on the

distribution of similarity score in genuine and im-

postor matching. Genuine matching compares two

ﬁngerprint templates from the same ﬁnger, whereas

impostor matching is for two ﬁngerprint templates

from different ﬁngers. The overall accuracy can be

illustrated by Receiver Operation Characteristics

(ROC) curve, which shows the dependence of

False Non-match Rate (FNMR) on False Match

Rate (FMR) at all thresholds. A series of indicators

are adopted to quantify the accuracy containing

Equal Error Rate (EER – the point where FNMR

and FMR yield the same value), FMR100 (the low-

est FNMR for FMR <= 1%), FMR1000 (the lowest

FNMR for FMR <= 0.1%) ZEROFMR (the lowest

FNMR for FMR <= 0%), and ZEROFNMR (the

lowest FMR for FNMR <= 0%) [15, 16]. Accuracy

usually attracts most of the attention in common

applications, bu t the algorithms cannot just be

characterized by these indicators.



Resource consumption can be measured through

three aspects: the amount of storage, time, and

memory required by the algorithms. The storage

cost is measured by the average/maximum size

of template for each database. The efﬁciency is

indicated by the average/maximum time in genu-

ine/imposter matching and the memory require-

ment is measured by the average/maximum size

of allocated memory in genuine/imposter match-

ing. The variation of the indicators through the

whole database reﬂects the stability of the tested

algorithms.



The scalability reﬂects the degradation of the accu-

racy with the growing scale of database, w hich is

available in one -to-many matching. It should be

evaluated on diffe rent-scale databases through

observing the relationship between the aforesaid

indicators and the scale of database. The interna-

tional competition FpVTE2003 [17] adopted three

different-scale databases to evaluate the scalability

of the tested algorithm.



Sensor interoperability denotes the ability to han-

dle the templates obtained from different sensors

[18]. The features in templates are sensitive to

different characteristics of multiple sensors in a

ﬁngerprint system. The comparison of measures

(accuracy, efﬁciency) between intra-sensor match-

ing (comparing tem plates from the same sensor)

and inter-sensor matching (comparing templates

from different sensors) somewhat reﬂects the inter-

operability of the matching approach. Research on

sensor interoperability is at its ﬂedgling stage and

so far there have been no authorized databases or

indicators for quantiﬁed evaluation.

Performance during evaluation is relative to many

objective conditions. Accuracy is inﬂuenced by both

the characteristics of database (size, average quality,

distortion) and the testing

▶ protocol, while resource

consumption relies on the hardware capability, so it

is meaningless to evaluate the matching approach

without considering these conditions. An authentic

evaluation should be conducted on the databases that

have independent training/testing parts and sufﬁcient

ﬁngerprints, and calculate the statistical indic ators

with a reasonable protocol. Because the above indica-

tors are statistical results, it should be reported how

believable the evaluation of these statistics really are.

The problem can be addressed by computing the con-

ﬁdence intervals on the distributio n of these values

[19]. The accuracy of these conﬁdence interval esti-

mates is ascertained by both correct estimation strate-

gies and correct dataset sampling.

The comparison of performance among various

matching algorithms is always a controversy. Dif -

ferent algorithms have different advantages and dis-

advantages; therefore it is unfair to directly conclude

one better than the other. Some research displays

experiment results conducted on the proprietary

databases using different protocols. This makes

it difﬁcult to compare the performance fairly.

Evaluating and comparing these indicators among

different algorithms is required to operate on the

500

Fingerprint Matching, Automatic

same public databases with the same authoritative

protocol and testing environment, such as FVC

and FpVTE.

Application

In AFIS, the ﬁngerprint matching can be applied in

two distinct models: veriﬁcation and identiﬁcation.

The veriﬁcation model is a one-to-one matching

(1:1) in which a user states his/her identity by means

of an ID and proves it with a ﬁngerprint. A new

ﬁngerprint sample taken from the user is compared

with the user’s previously registered or stored ﬁnger-

prints. The comparison only occurs once between the

input ﬁngerprint image and the selected sample from

the database following the claim of the user. If the

ﬁngerprints are successfully matched, the user is ver-

iﬁed as who he/she is claiming to be, and granted all

the privileges and access of the stated user. On the

contrary, the identiﬁcation process is a one-to-many

matching (1:N), in which a user need not claim his or

her identity. A new impression is taken from the user

and compared to the existing ﬁngerprints of registered

or stored users in the databases. The identiﬁcation can

be implemented with a sequence of veriﬁcation be-

tween the input template and the query templates

in the database. Fingerprint identiﬁcation requires

searching the database for a matched template or sev-

eral cand idates, which is a process more complex than

veriﬁcation. Although satisfactory performances have

been reported for ﬁngerprint veriﬁcation, both the

efﬁciency and accuracy of identiﬁcation deteriorate

seriously by simple extension of a 1:1 veriﬁcation pro-

cedure to a 1:N identiﬁcation system. It is still necessary

to improve the performance of ﬁngerprint matching in

the large-scale ﬁngerprint database. Fingerprint classi-

ﬁcation and indexing techniques are proved effective to

narrow down the searching space of veriﬁcation, which

will speed up the identifying process.

Different kinds of applications focus on different

requirements. For the same algorithm, the matching

threshold can be modulated or other parameters

conﬁgured to realize trade-off among these perfor-

mance indicators. There exists a strict relationship

between accuracy and resource consumption, FMR

and FNMR of each algorithm. For instance, both

FMR and FNMR are actually the functions of match-

ing threshold. The decrease in value makes the

algorithm more tolerant to the transformations

(lower FNMR), but increases the possibility of incor-

rectly matching two templates from different ﬁngers

(higher FMR). Contrarily, if the value increases, the

algorithm performs with higher FNMR and lower

FMR. According to the given application, the threshold

is carefully chosen as suitable for the special require-

ment. It is difﬁcult to develop a matching approach

omnipotent in every scenario, therefore s different

applications may at times need different algorithms.

The embedded applications (mobile phone, identit y

card) emphasize limited resources and put signiﬁcant

strain on the recognition reliability, because high per-

formance ﬁngerprint matching approaches tend to be

computationally intensive. In this case, we tend to

adopt these algorithms w ith lower resources consump-

tion. In contrast, the resource-unlimited applications

equipped with adequate resources attach more impor-

tance to accuracy rather than the computation and

storage expenditure. For instance, ﬁngerprint match -

ing in network security operates on the distributed

computer system with a ‘‘Trustworthy Authority + Re-

mote Client’’ mode, where extreme accuracy is the

most crucial target. In these situations, we choose the

algorithms that have more accuracy despite of the

possible computational complexity.

Summary

Recently, there have been great advances in the research

on automatic ﬁngerprint matching. However, the vari-

ous applications of AFIS in pers onal identiﬁcation

desire further improvement of the performance of

matching algorithms. Recent research demonstrates

that fusion in different levels (feature, score, decision)

is effective in improving the performa nce in many

aspects, attracting increasing interest. Besides the fea-

tures, the fusion of multiple independent matchers

[20, 21 ] is likely to ameliorate the accuracy of ﬁnger-

print matching.

Related Entries

▶ Fingerprint Classiﬁcation

▶ Fingerprint Clustering

▶ Fingerprint Matching, Manual

▶ Fingerprint Recognition

▶ Fingerprint Templates

Fingerprint Matching, Automatic

501

References

l. Maltoni, D., Maio, D., Jain, A.K., Prabhakar, S.: Handbook of

Fingerprint Recognition. Springer, Berlin (2003)

2. Ross, A., Jain, A.K.: Biometric sensor interoperability: a case

study in ﬁngerprints, BioAW 2004. LNCS 3087, 134–145 (2004)

3. Bazen, A.M., Gerez, S.H.: An intrinsic coordinate system for

ﬁngerprint matching. In: Third International Conference on

Audio- and Video-Based Biometric Person Authentication,

Halmstad, Sweden (2001)

4. Luo, X.P., Tian, J., Wu, Y.: A Minutia matching algorithm in

ﬁngerprint veriﬁcation. Fifteenth ICPR 4, 833–836 (2000)

5. Jain, A.K., Lin, H., Bolle, R.: On-line ﬁngerprint veriﬁcation. IEEE

Trans. Pattern Recogn. Machine Intell. 19(4), 302–314 (1997)

6. Kovacs-Vajna, Z.M.: A ﬁngerprint veriﬁcation system based on

triangular matching and dynamic time warping. IEEE Trans.

Pattern Recogn. Machine Intell. 22(11), 1266–1276 (2000)

7. Chen, X.J., Tian, J., Yang, X.: A new algorithm for distorted

ﬁngerprints matching based on normalized fuzzy similarity

measure. IEEE Trans. Image Process. 15(3), 767–776 (2006)

8. Bazen, A.M., Gerez, S.H.: Fingerprint matching by thin-plate

spline modelling of elastic deformations. Pattern Recogn. 36(8),

1859–1867 (2003)

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

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

10. He, Y.L., Tian, J., Li, L., Yang, X.: Fingerprint matching based

on global comprehensive similarity. IEEE Trans. Pattern Analy.

Machine Intell. 28(6), 850–862 (2006)

11. Jain, A.K., Chen, Y., Demirkus, M.: Pores and ridges: high-

resolution ﬁngerprint matching using level 3 features. IEEE

Trans. Pattern Recogn. Machine Intell. 29(1), 15–27 (2007)

12. Jain, A.K., Prabhakar, S., Lin, H., Pankanti, S.: Filterbank-based

ﬁngerprint matching. IEEE Trans. Image Process. 9(5), 846–859

(2000)

13. Gu, J., Zhou, J., Yang, C.: Fingerprint recognition by combining

global structure and local cues. IEEE Trans. Image Process 15(7),

1952–1964 (2006)

14. Wang, X.C., Li, J.W., Niu, Y.M.: Fingerprint matching using

OrientationCodes and PolyLines. Pattern Recogn. 40(11),

3164–3177 (2007)

15. http://bias.csr.unibo.it/fvc2002/perfeval.asp

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

Performance evaluation of ﬁngerprint veriﬁcation systems. IEEE

Trans. Pattern Recogn. Machine Intell. 28(1), 3–18 (2006)

17. http://fpvte.nist.gov

18. Ross, A., Jain, A.K.: Biometric sensor interoperability: a

case study in ﬁngerprints. Proc. Biometric Authentication:

ECCV 2004 International Workshop 3087, 134–145 (2004)

19. Bolle, R.M., Ratha, N.K., Pankanti, S.: An evaluation of error

conﬁdence interval estimation methods. Fifteenth ICPR 3,

103–106 (2004)

20. Ross, A., Jain, A.K., Reisman, J.: A hybrid ﬁngerprint matcher.

Pattern Recogn. 36(7), 1661–1673 (2003)

21. Bhakar, S., Jain, A.K.: Decision-level fusion in ﬁngerprint

veriﬁcation. Pattern Recogn. 35(4), 861–874 (2002)

22. Chen, X.J., Tian, J., Yang, X. An algorithm for distorted ﬁnger-

print matching based on local triangle features set. IEEE Trans.

Inform. Forensics Security 1(2), 169–177 (2006)

Fingerprint Matching, Manual

HER MAN BERGMAN

,ARIE ZEELENBERG

Certiﬁed Fingerprint Expert, San Francisco, USA

Senior Advisor Fingerprints, National Police Force,

The Netherlands

Synonyms

Identiﬁcation; Individualization; Minutial

Definition

Identiﬁcation has been deﬁned as the determination by

a ﬁngerprint examiner that two examined images of

friction ridge skin are deposited by the sam e source

(ﬁnger, palm or foot), with the goal of determining the

identity of a donor. If this can be established it is

generally accepted within the discipline that given the

uniqueness or

▶ individuality of friction ridge skin,

this ﬁngerprint can be attributed to this donor at the

same time excluding all others. (In this contribution

an expert for practical reasons is referred to as ‘‘he’’.

Female experts should not feel excluded but may

comfort themselves with the idea that with respect to

erroneous identiﬁcations also the male form is used)

Fingerprint Matching: Manual

The matching process described here applies to marks or

latent prints found at a crime scene or on pieces of

evidence associated with a crime. Those marks tend to

be incomplete and of lesser quality than

▶ comparison

prints. The process where known prints are compared,

one to one or one to many, to verify an identity has

become an increasingly automated process. Because of

the amount of quality and quantity of data available

and the accuracy of current Automated Fingerprint

Identiﬁcation Systems (AFIS) this process can be ap-

plied in a ‘‘lights out’’ mode or monitored by examiners.

This automated process to determine individuality

is generally referred to as ‘‘matching’’ and is executed

by matching algorithms. For the process where latent

prints or marks are analyzed and compared by an

examiner the more generic term identiﬁcation or indi-

vidualization is used rather than matching.

502

Fingerprint Matching, Manual

The identiﬁcation process is a one to one com-

parison and starts after a similar print is found which

cannot be excluded as being the same at face value.

Three possible scenarios can lead to this:

A candidate can be the result of an AFIS search in

which the similarity of the extracted features is calcu-

lated against known exemplars in a digital repository.

If one of the best resembling candidates cannot

be excluded it might be eligible for input in an identi-

ﬁcation process.

Second, a candidate can be selected after manual

comparison of one or more named suspects.

Third, a candidate may be found through a manual

search of a physical ﬁngerprint repository. This last

occasion becomes increasingly rare because physical

ﬁngerprint repositories and manual searching become

distinct by the broad use of AFIS.

The process by which the expert examines possible

candidates focuses more on elimination based on dif-

ferences than on weighing of similarities . At this stage

the examiner searches for differences in the overall

pattern formed by the ridges which is considered the

ﬁrst of three levels of information that are generally

distinguished [1]. They are addressed to as

▶ the ﬁrst,

▶ second and ▶ third level detail.

When an expert manually compares a mark against

known, or comparison prints he visually assesses the

main aspects of the ridge ﬂow and/or a discernible

pattern and a chosen target group of

▶ minutiae which

he can relate to a recognizable area or location in the

mark such as a delta, core or along the type lines.

This information is used to eliminate compared

prints, this exclusion may be a very fast process. At

one glance an expert may see that a compared donor

shows 10 whorl patterns in the ﬁngertips while he is

looking for a loo p. Even so a donor with a number of

loops to the right with high ridge counts between the

delta and core can be excluded deﬁnitively if the

mark has a low ridge count. If no exclusion on ridge

ﬂow is possible because it is similar the remaining

print will be compared keeping the target group in

mind and looking for differences in the known print

at the given positions relative to known locations. If

he initially ﬁnds small clusters in a similar sequence he

will then expand the assessed area both in the mark

and the known exemplar.

If the print does not originate from the same source

he will quickly ﬁnd discrepancies and the comparison

print will be excluded. If exclusion fails, the candidate

will be incl uded in the identiﬁcation process.

The identification process

The generally accepted methodology for the identiﬁ-

cation process of friction ridge impressions is known as

ACE-V [1] or a variation of this [2]. ACE-V is the

acronym for Analysis, Comparison and Evaluation

followed by Veriﬁcation by another expert. ACE-V

was ﬁrst introduced by R.A. Huber [3] and later by

D. Ashbaugh [1] for the examination of friction ridge

skin. This methodology is generally accepted in foren-

sics as a universal protocol to promote reproducibility

and objectivity and should allow for the validation of

the stated conclusions by reference to the process

through which they are constructed.

It has been argued that ACE-V may not fully pro-

vide the requirements [4] necessary for an identiﬁca-

tion technique which should be explicit and deﬁned in

more detail [5]. Professors van Koppen and Crombag

[6, 7] proposed the use of a descriptive model and a

decision making model in forensic identiﬁcation of

ear-, lip- and ﬁngerprints.

The Interpol European Expert Group on Finger-

print Identiﬁcation (IEEGFI) report [2] not only

describes a method similar to ACE-V (Fig. 1), but

also provides both a descriptive model and a decisio n

making model [6].

These models present a common terminology,

grounds for establishing the value of features, rules of

thumb, describe the pitfalls and provide good guidance

for decision-making with respect to details and the

overall decision of identiﬁcation. It is essenti al to re-

produce the whole process rather than to conﬁne re-

producibility to the conclusion.

The IEEGFI uses the word information phase as a

synonym for the analyses phase and addresses the

evaluation as the balance phase.

Analyses

A thorough and objective analysis of the latent print is

the basis of a sound process, an unbiased establishment

of the quantity and quality of available data is the aim.

The analysis is the establishment of features and their

properties and values recorded in a combination of

mental and explicit written note s of all observed data.

A copy of the image of the latent can be marked up in

order to document observations. All three levels of

information that are regarded as properties of friction

ridge skin are assessed to determine their reliability

Fingerprint Matching, Manual

503

and value, taking into account the inﬂuence of devel-

opment technique(s) used, the exhibit, distortion,

surface, deposition pressure, matrix, and anatomical

aspects. Ambiguous Galton features of which the exact

location cannot be seen at face value can still be estab-

lished by

▶ tracing. In these instances the ridge detail

and the exact appearance of the detected feature are

unknown and may add little weight to the value of

the latent and, subsequently, to the comparison. Nev-

ertheless, it can be helpful to check whether certain

Galton features in the comparison print are at least

not in conﬂict with the latent.

Although a good practise in all cases, it is acknowl-

edged that not all latent prints require such an in-

depth a nalysis. In instances of high quality latent prints

with unambiguous and/or an abundance of data, the

analysis can be very quick.

However, it should be stressed that with low quality

and quantity latent prints a full in-depth analysis is

essential. The importance and depth of the analysis is

inversely proportional to the quality of the latent print.

The IEEGFI II proposed a special procedure,

‘‘

▶ The need for a questionable ID procedure’’ for

complex examinations [6]. The examiner has to form

an opinion about the quality, quantity, and reliability

of the observed data in the latent print and on the

basis of this he has to decide whether the latent

print has sufﬁcient potential to relate it to its unique

source. If that is the case he moves on to the compari-

son phase.

Comparison

The latent and the comparison prints are placed side by

side enabling accurate comparison and the preserva-

tion of observations.

The data obtained in the analysis phase form the

basis and guide for the comparison process and should

be leading. During comparison not only data in the

latent are checked against the comparison print, but

also data found in the comparison print are cross-

checked whether or not they are present in the latent.

The relations of all features within the conﬁgura-

tion are checked through triangulation [8]. This is

done by following the ridges or furrows and counting

the number of intervening ridges between features

along a virtual connecting line. The relative location

aspects and relations to other features in the latent have

to be with in tolerance compared to the features in the

corresponding locations of the comparison print. (The

direction in which the neighboring feature is found is

Fingerprint Matching, Manual. Figure 1 Diagram ACE.

504

Fingerprint Matching, Manual

checked towards the general ridge ﬂow and relative to

the connecting line with other minutiae in the same

area.) Due to the ﬂexibility of the skin the interrela-

tionship of features can be disturbed, but as in the case

of a stretched spider web the relative positions remain

the same (Fig. 2).

At this stage the expert also looks for similar third

level detail which he relates to the location of second

level detail.

It can be a very powerful contribution to each

individual minutia and to the whole of a print but its

accurate representation is dependent on a large num-

ber of variables such as pressure, moisture, the surface,

and the detection technique. Reliable third level detail

in latent prints is a gift rather than a given fact. It is

often difﬁcult to draw a distinction between third level

detail and anomalies. Matching third level detail is not

very common and often calls for rationalization. In-

stead, the relationship of the minute events amongst

them and with minutiae is more often studied. A large

pore on the edge of a line followed by a small one in the

centre of a ridge, the ﬂow of an individual ridge like a

recognizable river bed, a small dot lying in front of a

tapered ridge ending are examples of ridge detail that,

if similar, can be very signiﬁcant contributions to the

weight of the comparison (Fig. 3).

Fingerprint Matching, Manual. Figure 2 Triangulation/Distortion.

Fingerprint Matching, Manual. Figure 3 Third level detail of exceptional quality, organic shapes that are found in a

similar fashion on exact corresponding locations.

Fingerprint Matching, Manual

505

Overall similarities should be apparent and demon-

strable and be primarily based upon ﬁndings obtained

in the analysis phase. Thus, avoiding the implementa-

tion of features found in the sup posed original into

the latent. When marking corresponding features

it is important to establish the existence, the relations

and their signiﬁcance. For each individual point of

similarity the qu ality may differ. If a point is clear

and shows corresponding ridge detail its value is sig-

niﬁcantly higher than points that do not have these

properties.

Dissimilarities and/or discrepancies should be

detected, assessed, noted and accounted for. Any ex-

planation of dissimilarities should preferably reﬂect

the observations made during the analysis phase. An

opinion has to be formed whether the differences in

appearance are considered distortions or discrepancies.

In the case of discrepancies, the conclusion should be

an exclusion or/and inconclusive.

There is a distinct difference between the com-

parison of minutiae and ridge detail. Minutiae must

be the same and ridge detail can be the sam e. Whereas

the basic properties of the Galton points are ﬁrmly

established during the analysis phase true third level

detail is often only acknowledged and conﬁrmed dur-

ing comparison taking the supposed original as

the blueprint. This carries the risk of a picking attitude

of the expert who may select everything that appears

to be similar and ignore all that is not.

Further, this promotes the risk of circular reasoning

[6] or ‘‘gestalt analyses’’ [4], instead of proving origin

by the similarities one ‘‘proves’’ similarities by the

assumed origin.

It has been discussed that the ACE-V protocol is a

recurring and reversible process [9 , 10]. Opinions vary

however whether or not the process should be totally

recurring, and reversible (or up to a certain level) or

that attempts should be made to conﬁne it to a more

linear process wherever possible. With a recurring

and reversible process the risk of inserting information

of the ‘‘known’’ exemplar into the unknown is higher

than in a strict linear process in which ACE-V is exe-

cuted once in the exact order.

The risk of making a (subconscious) decision early

in the comparative process and the potential inﬂuence

of it must be recognized [11]. The comparison must be

an unbiased ‘‘step by step’’ building process ensuring

that the data in the latent and comparison print match,

with nothing in disagreement w hich cannot be

logically explained and accounted for. The decision

must be made at the end of the process only.

An expert who has executed the process of search-

ing and elimination has performed an initial and in-

complete analysis directed towards elimination and/or

the search process. Since he has singled out a compari-

son print for the identiﬁcation process he has arrived

at a preliminary conclusion about possible identiﬁca-

tion. With an eye to the ‘‘half baked’’ analys is and the

preliminar y conclusion it is advisable that the expert

renounces himself from the identiﬁcation process.

Evaluation and Preliminary

Conclusion

Requirements for the conclusion of identiﬁcation

as provided by SWGFAST [12] are; agreement of sufﬁ-

cient friction ridge detail; determined by a competent

examiner; applied to a common area in both impres-

sions; based on quantity and quality of friction ridge

detail; without any discrepancy and a reproducible con-

clusion. The total volume in agreement is a composition

of coherent qualitative and quantitative information.

In the USA, after a 3-year study by a Standardiza-

tion Committee, the use of a numerica l standard

was discouraged by the adoption of a resolution at a

conference of the International Association for Identi-

ﬁcation which stated: ‘‘no scientiﬁc basis exists for

requiring that a predetermined minimu m of friction

ridge features must be present in two imp ressions in

order to establish positive identiﬁcation’’.

Sufﬁciency is since left to the discretion of the expert

and measured against his training, knowledge and expe-

rience, and his personal standard. SWGFAST [12]relates

reproducibility primarily to the format of ACE-Vand to

the conclusion. This position is known as the Expert

Opinion System or the holistic approach [2, 13].

In many other countries a numerical standard is

used as an aid to measure sufﬁciency which is called

the Empirical Standard Approach [2, 13]. This stan-

dard expresses a minimum number of minutiae in

agreement that is used as a common, empirical refer-

ence and a tool to guide the process, to facilitate

veriﬁcation and to obtain and guarantee quality.

In either system if an expert decides that in his

opinion identiﬁcation is justiﬁable because the equa-

tion is both sufﬁcient and cogent he will put it up for

veriﬁcation [14].

506

Fingerprint Matching, Manual

Verification

The postulated conclusion should be reproducible by

another examiner applying the same methodolog y.

This is accomplished by the veriﬁcation phase of the

ACE-V methodology.

The reliability of a conclusion can be che cked and

demonstrated by an independent veriﬁcation. Veriﬁca-

tion can be limited to another expert independently

arriving at the same conclusion or by repeating

and checking the whole examination of the initial

examiner. The veriﬁcation process should have the

characteristics of scrutiny rather than conﬁrmation of

the conclusion. (Also see mistakes.)

If the veriﬁer is satisﬁed that the process and the

conclusion meet the requirements, then the conclusion

is conﬁrmed and the identiﬁcation is established.

Conclusion

The conclusion of identiﬁcation is a veriﬁed opinion

that the investigated latent and the comparison print

come from the same source. It also implicates the expec-

tation of reproducibility, i.e., any other examiner using

the same methodology should arrive at the same con-

clusion. Given the empirical, biological and statistical

support for friction ridge skin uniqueness or individual-

ity, an identiﬁed ﬁngerprint is attributed to a single

donor [15].

Charting

The use of a computer screen during the analysis and

side by side comparison of friction ridge images can

be of tremendous help in the examination process. The

data and the relations of the conﬁguration can be

cross-checked, in par ticular with ambiguous informa-

tion. Details can be better observed and compared by

enlarging and/or enhancement of the images to opti-

mize the perceptibility of the characteristics in print.

This not only increases the quantity and accuracy of

the data observed [16], but also makes it easier to value

and appreciate the similarities and dissimilarities.

At the same time similarities can be mar ked up,

printed and saved for documentation purposes.

In order to meet the requirement of demonstrabil-

ity of all the phenomena upon which the expert bases

his ﬁndings and conclusion this tool is indispensable.

It also facilitates consultation and discussion amongst

experts.

Historically, court charts have been produced in

which coinciding minutiae have been marked up and

numbered. Court charts can be a useful tool to dem-

onstrate some of the ﬁndings but are just meager

illustrations of a very complex process and should

not be taken as ultimate proof. (The simple argument

for that is the fact that in the past with erroneous

identiﬁcations court charts were produced with even

extensive numbers of marked similarities (Fig. 4).)

Mistakes

Error rates, an endless source of scientiﬁc debates, phi-

losophies and semantics, will not be covered nor decided

here but known errors will be discussed. It is obvious

that, in relation to the immense numbers of identiﬁca-

tions effected over the more than hundred years of

ﬁngerprinting, the number of erroneous identiﬁcations

that surfaced is extremely low. In a study by Simon Cole

[17] 22 erroneous identiﬁcations were investigated for

the period from 1920 to 2004. Even if the number is

tripled or multiplied by 10 incorporating a number

of dark ﬁgures the positive ratio against the millions

of identiﬁcations performed remains. Some support

for this positive ratio is also found in data collected

during comparison training exercises [18].

This does not implicate that mistakes are regarded

as part of the system and inevitable, on the contrary.

Every mistake is one too many and can do irreparable

harm to innocent people. The profession should take

all possible measures to prevent them.

Another ongoing debate is whether the mistakes can

be attributed to ﬂaws in the technique or to human

error or whether the two can be separated at all. How-

ever, it is clear that erroneous identiﬁcations are dis-

covered and exposed by experts. This is a strong

indication that the human factor is dominant. Second,

it is important to note that the examination of ques-

tioned identiﬁca tions can be repeated and checked

endlessly. When opinions differ upon sufﬁciency

the comparsion should be regarded as inconclusive.

In general, experts view mistakes very seriously. They

believe that making a mistake is the worst thing that can

happen to them and may discredit them in the eyes of

their contemporaries.

Fingerprint Matching, Manual

507

The paradox is that the acceptance of the sus-

ceptibility for human error by experts should be the

basis for a quality system, whilst very often the initial

response of experts to such criticism is defensive rather

than open minded. This blocks the feedback essential

for a quality cycle.

A preliminary analysis of mistaken identiﬁcations

revealed the following factors:



It concerned a border line latent with respect to

quality, quantity or both.



There was no apparent relationship between the

organization or level of experience of the exper t

(s) involved.



Veriﬁcation was degenerated to conﬁrmation

rather than scrutiny.



Experts were biased by domai n irrelevant

information.



Discrepancies were ignored or erroneously attrib-

uted to distortion.



Applied tolerances were too wide given the quality

of the latent. This is another paradox; ‘‘the worse the

print the larger the tolerances applied’’; experts may

attribute differences to the lack of quality and dis-

tortion and ‘‘explain them away’’ something they

may not do with an image of good quality. Thus,

bad quality may not only conceal real discrepancies,

but also provide an excuse for it at the same time.

In general, there is a growing opinion that a number of

psychological factors may potentially contribute to

cognitive and decision-making errors [19].

Examples are; the primacy effect, when informa-

tion is judged in the light of an early opinion; and

conﬁrmation biases like myside bias and truth bias

[11] are found in all types of ﬁelds as well as in

ordinary life. One major concern is that sufﬁciency

may be established after the comparison process and

as such after a conscious or subconscious decision is

made about identity. This makes the expert more vul-

nerable to bias [19].

Studies have been done to enhance insight into the

potential inﬂuences of bias during the examination of

ﬁngerprints [20].

Infallible or Reliable?

Some have criticized the profession for the explicit or

implicit claim of infallibility [17, 21].

The apparent reliability of ﬁngerprint identiﬁca-

tion for decades may have created this image as

reﬂected by the proverbial expression ‘‘as reliable as a

ﬁngerprint’’. This meant an image so strong that all

other forensic techniques were compared against it,

much like the introduction of DNA that was errone-

ously labeled the ‘‘genetic ﬁngerprint’’.

Responsible experts never claim infallibility be-

cause this is an unsustainable and unscientiﬁc position.

In retrospect, however, ﬁngerprints in general can

claim a record of great reliability, but as in any human

endeavor mistakes occur so safeguards have to be in

place.

Fingerprint Matching, Manual. Figure 4 Example of a court chart with 21 coinciding Galton characteristics marked.

508

Fingerprint Matching, Manual

The main ground for quality is the acceptance of

fallibility by individuals and communities. With that in

mind, instruments to achieve a solid conclusion,

the rigorous application of the methodology, Quality

Assurance protocols, training, testing and transpar-

ency, will be applied and maintained with conviction

and can be further improved.

Per individual case reliability of a conclusion can be

reached and demonstrated by veriﬁcation, peer review

and counterchecks by independent experts. This pro-

cess can be repeated over and over again without

affecting the material.

Related Entries

▶ Classiﬁcation

▶ Feature Extraction

▶ Fingerprint Classiﬁcation

▶ Fingerprint Matching Automatic

▶ Individuality

References

1. Ashbaugh, D.: Quantitative-Qualitative friction ridge analysis,

p. 105. CRC Press, Boca Raton, FL (1999)

2. Method for Fingerprint Identiﬁcation, Interpol European

Expe rt Group for Fingerprint Identiﬁcat ion Report I http://

www.interpol.int/ Public/Forensic/ﬁngerprints/WorkingPartie s/

default.asp

3. Huber, R.A.: Expert witness. Criminal Law Quarterly. 2, 276–295

(1959)

4. Rudin, N., Inman, K.: The proceedings of lunch The CAC

News of the California Association of Criminalists, 2nd Quarter

(2005) http://www.cacnews.org/

5. A Review of the FBI’s Handling of the Brandon Mavﬁeld Case,

Ofﬁce of the Inspector General, http://www.usdoj.gov/oig/

special/s0601/PDF_list.htm pp. 7, 198–199. (2006)

6. Method for Fingerprint Identiﬁcation, Interpol European

Expert Group for Fingerprint Identiﬁcation Report II. http://

www.interpol.int/ Public/Forensic/ﬁngerprints/WorkingPartie s/

default.asp

7. van Koppen, P.J., Crombag, H.H.: Over Oren, Lippen en

Vingers. Nederlands Juristenblad (2000)

8. Hare, K.: Proportional analysis: The science of comparison.

J. Forensic Ident. 53, 700 (2003)

9. Vanderkolk, J.R.: ACE-V: A Model. J Forensic Ident. 54 (2004)

10. Mc Kasson, S.C., Richards, C.A.: Speaking as an expert: A guide

for the identiﬁcation sciences from the laboratory to the court-

room, 131–138 (1998)

11. Nickerson, Raymond S.: Conﬁrmation bias: A ubiquitous phe-

nomenon in many guises. Rev. Gen. Psychol. 2, 175–220 (1998)

12. Scientiﬁc Working Group on Friction Ridge Analyses, Study

and Technology http://www.swgfast.org/Standards_for_Conclu

sions_ver_1_0.pdf

13. C. Champod, et al., Fingerprints and other ridge skin impres-

sions, 27–31 (2004)

14. Thornton, J.: ‘‘Setting Standards in the Comparison and Identi-

ﬁcation’’ In: 84th Annual Training Conference of the California

State Division of IAI Laughlin, Nevada, May 9 (2000) http://

www.latent-prints.com/Thornton.htm

15. Moenssens, A.A.: Is ﬁngerprint identiﬁcation a science? http://

forensic-evidence.com/site/ID/ID00004_2.html

16. Langenburg, G.M.: A statistical analysis of the ACE-V method-

ology: Analysis stage. J. Forensic Ident. 54 (2004)

17. Cole, S.A.: More than zero: Accounting for error in latent ﬁnger-

print identiﬁcation. J. Crim. Law Criminol. 95 (2005)

18. Wertheim, K., Langenburg, G. Moenssens, A.: ‘‘A report of latent

print examiner accuracy during comparison training exercises’’.

J. Forensic Ident. 56, 55–93 (2006)

19. Itiel, D., Charlton, D., Pe

ron, A.E.: ‘‘Why are experts prone to

error?’’ Forensic Sci. Int. 156, 74–78 (2006)

20. Schiffer B., Champod, C.: The potential (Negative) inﬂuence of

observational biases at the analysis stage of ﬁngermark individ-

ualization. Forensic Sci. Int. 167, 116–120 (2007)

21. Saks, M., Koehler, J.: The coming paradigm shift in forensic

identiﬁcation science. Science, 309, 892 (2005)

Fingerprint Pre-Matching

▶ Fingerprint Classiﬁcation

Fingerprint Quality

The intrinsic characteristic of a ﬁngerprint image that

may be used to determine its suitabilit y for further

processing by the biometric system or assess its con-

formance to pre-established standards is ﬁngerprint

quality. The quality of a biometric signal is a numerical

value (or a vector) that measures this intrinsic

attribute.

▶ Individuality of Fingerprints

Fingerprint Quality

509