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

Подождите немного. Документ загружается.

Fingerprint Reading

▶ Biometric Sample Acquisition

Fingerprint Recognition, Overview

DAV I DE MALTONI

DEIS, University of Bologna, Italy

Synonym

Fingerprint Biometric

Definition

Fingerprint recognition allows a person to be veriﬁed

or identiﬁed through the analysis and comparison of

his or her ﬁnger dermal ridges. Fingerprint recognition

was one of the ﬁrst techniques used for automatically

identifying people and today is still one of the most

popular and effective biometric techniques.

Introduction

A ﬁngerprint is the representation of the dermal ridges

of a ﬁnger [1]. Dermal ridges form through a combi-

nation of genetic and environmental factors; the genet-

ic code in DNA gives general instructions on the way

skin should form in a developing fetus, but the speciﬁc

way it forms is a result of random events such as the

exact position of the fetus in the womb at a particular

moment. This is the reason why even the ﬁngerprints

of identical twins are different [2]. Fingerprints are fully

formed at about 7 months of fetus development and

ﬁnger ridge conﬁgurations do not change throughout

the life of an individual, except in case of accidents such

as severe cuts on the ﬁngertips. This stability makes

ﬁngerprints a very attractive biometric identiﬁer. Several

mathematical models based on the

▶ anatomy of fric-

tion ridge skin were developed over the years to quan-

tify

▶ ﬁngerprint individuality [3] and to prove that

ﬁnding two persons with identical ﬁngerprints is ex-

tremely unlikely. This does not imply that ﬁngerprint

recognition is a perfect technique: in fact, various

kinds of errors can affect ﬁngerprint acquisition and

processing thus requiring to introduce thresholds to

decide if two ﬁngerprint impressions are similar

enough to be considered belonging to the same person.

As for any biometric technique, a sound performance

evaluation (see

▶ ﬁngerprint databa ses and evalua-

tion) is extremely important to estimate the accuracy

of a ﬁngerprint-based biometric system and to under-

stand if it is well-suited for a particular application.

Recent independent evaluation campaigns such as

FVC2006 [4] proved that state-of-the art ﬁngerprint

recognition algorithms are nowadays ver y accurate

(i.e., EER less than 0.1% for a database collected with

a large area optical scanner).

History

Human ﬁngerprints have been discovered on archaeo-

logical artefacts and historical items (Fig. 1). Although

these ﬁndings prove that ancient people used ﬁnger-

prints for a number of purposes, it was not until the

late sixteenth century that the modern scientiﬁc ﬁnger-

print studies were initiated [5]. In 1686, Marcello

Malpighi, a professor of anatomy at the University of

Bologna, noted the presence of ridges, spirals and

loops in ﬁngerprints. Henry Fauld, in 1880, was the

ﬁrst to scientiﬁcally suggest the individuality of ﬁnger-

prints based on an empirical observation. At the same

time, Herschel asserted that he had practiced ﬁnger-

print recognition for about 20 years. In the late nine-

teenth century, Sir Francis Galton conducted an

extensive study on ﬁngerprints; in 1888 he introduced

the

▶ minutiae features for ﬁngerprint matching. An-

other important advance was made in 1899 by Edward

Henry, who established the well-kn own ‘‘Henry sys-

tem’’ of

▶ ﬁngerprint classiﬁcation.

In the early twentieth century, ﬁngerprint recogni-

tion was formally accepted as a valid identiﬁcation

method and became a standard routine in forensics

[5]. Fingerprint identiﬁcation agencies were set up

worldwide and criminal ﬁngerprint databases were

established; for instance, the FBI ﬁngerprint identiﬁca-

tion division was set up, in 1924, with a database of

810,000 ﬁngerprint cards. With the rapid expansion

of ﬁngerprint recognition in forensics, operational

510

Fingerprint Reading

ﬁngerprint databases grew so large that manual ﬁnger-

print identiﬁcat ion (see

▶ ﬁngerprint matching, man-

ual) became infeasible; for example, the total number

of ﬁngerprint cards in the FBI ﬁngerprint database

stands well over 200 million and is continuously grow-

ing. With thousands of requests being received daily,

even a team of more than 1300 ﬁngerprint experts were

not able to provide timely responses to these requests.

Starting in the early 1960s, the FBI, Home Ofﬁce in the

UK, and Paris Police Depar tment began to invest

a large amount of effort in developing Automati c

Fingerprint Identiﬁcation Systems (

▶ AFIS). Based on

the observations of how human ﬁngerprint experts per-

form ﬁngerprint recognition, three major problems in

designing AFIS were identiﬁed and investigated: digital

ﬁngerprint acquisition, local ridge feature extraction,

and ridge characteristic pattern matching. Their efforts

were so successful that today almost every law enforce-

ment agency worldwide uses an AFIS. These systems

have greatly improved the operational productivity of

law enforcement agencies and reduced the cost of hiring

and training human ﬁngerprint experts.

Automatic ﬁngerprint recognition technology

has now rapidly grown beyond forensic applications.

On the one side, together with face, ﬁngerprint is the

main biometric modality for electronic documents

(e-passport, visas, ID cards, etc) used to enforce border

crossing and citizen security. On the other side, thanks

to a ver y good performance/cost tradeoff, ﬁngerprint-

based biometric systems are becoming very popular

and are being deployed in a wide range of commercial

applications such as logon to computers and networks,

physical access control, ATMs.

Components of a Fingerprint

Recognition System

The block diagram of a ﬁngerprint-based recognition

system is depicted in Fig. 2.

A ﬁngerprint is acquired through a live-scan

▶ fin-

gerprint device that allows to simply and quickly cap-

ture a digital ﬁngerprint image: most of the ﬁngerprint

devices sample the pattern at 500 DPI (Dots per Inch)

and produce an 8-bit gray-scale raw image (see Fig. 3).

Some devices also include fake detection mechanisms

(see

▶ ﬁngerprint fake detection) that allow to reveal

spooﬁng attacks carried out with fake ﬁngers.

The acquired raw image is then passed to a quality

control module that evaluates if the ﬁngerprint sample

quality is good enough to correctly process it and to

extract reliable features. In case of insufﬁcient quality,

the system rejects the sample and invites the user to

repeat the acquisition; otherwise, the raw image is passed

to an

▶ image enhancement module whose goal is

improving the clarity of the ridge pattern, especially in

noisy region, to simplify the subsequent feature extrac-

tion. Special digital ﬁltering techniques, known as con-

textual ﬁltering [1], are usually adopted at this stage; the

output enhanced image can still be a gray-scale image or

become a black-and-white image. The

▶ feature extrac-

tion module further processes the enhanced image and

extracts a set of features from it. This feature set often

includes minutiae but, depending on the matching algo-

rithm, other features (e.g., local orientation, local fre-

quency,singularities,ridgeshapes,ridgecounts,partsof

the enhanc ed image, etc.) can be extracted in conjunc-

tion with (or instead of) minutiae.

Fingerprint Recognition, Overview. Figure 1 Examples of archaeological fingerprint carvings and historic fingerprint

impressions: (a) Neolithic carvings (Gavrinis Island); (b) standing stone (Goat Island, 2000 B.C.); (c) an impression on a

Palestinian lamp (400 A.D.). Figures courtesy of A. Moenssens and R. Gaensslen.

Fingerprint Recognition, Overview

511

Finally, the ﬁngerprint matching module (see ▶ fin-

gerprint matching, automatic) retrieves from a system

database one or more templates (see

▶ ﬁngerprint

templates) and matches it/them with the features

extracted from the current sample. Most of the match-

ing algorithms, following the well established manual

method (see

▶ ﬁngerprint matching, manual), com-

pare two ﬁngerprints by searching the spatial correspon-

dence of a minimum number of minutiae; this is not a

simple task because of the large variations (e.g., dis-

placement, rotation, skin condition, distortion, noise,

etc.) that can characterize two ﬁngerprint images ac-

quired from the same ﬁnger at different times. If the

systems is operating in veriﬁcation mode, the user has

been required to claim his identity and therefore just

one template is retrieved from the database and

matched with the current sample; if the system is

operating in identiﬁcation mode the current sample is

matched against all the database templates to check is

one of them is sufﬁciently similar.

Protecting ﬁngerprint templates is very important

to avoid attacks to ﬁngerprint-based biometric systems

[6] and to preserve user privacy: cryptography techni-

ques can be used to this purpose (see

▶ Fingerprints

Hashing).

Large-Scale Automatic Fingerprint

Identification Systems

Large-scale automatic ﬁngerprint identiﬁcation sys-

tems (AFIS) are used in forensic and civil government

applications. The basic functioning of these systems

is the same as described in the previous section, but

a number of ad-hoc optimizations are employed to

effectively and efﬁciently store, retrieve and match

millions of ﬁngerprints in a few seconds. In the past,

special dedicated hardware and stor age devices were

used to guarantee the required throughput; nowadays,

most of the AFIS cores run on conventional hardware

(e.g., cluster of personal computers) and the software is

the main responsible of the system efﬁciency. Finger-

print classiﬁcation and

▶ ﬁngerprint indexing are the

two main techniques used to speed-up a ﬁngerprint

search in a large database [1]. The former allows to split

the database in a number of partitions and to limit the

search to the partition to which the searched sample

belongs to. The latter enables sorting the database tem-

plates according to the similarity with the searched

Fingerprint Recognition, Overview. Figure 2 Block diagram of a fingerprint-based recognition system.

Fingerprint Recognition, Overview. Figure 3 Example of

fingerprint images from FVC2006 databases [4].

512

Fingerprint Recognition, Overview

sample, so that the probability to ﬁnd a mate in the ﬁrst

attempts increases signiﬁcantly. Even if the capacity of

mass storage devices is continuously growing, storing

ﬁngerprints as uncompressed raw images would require

too much space (nowadays AFIS must store billions of

ﬁngerprint images) and would increase the time neces-

sary to transmit a ﬁngerprint record over a network; to

alleviate this problem, without compromising recogni-

tion accuracy, speciﬁc

▶ ﬁngerprint compression tech-

niques such as WSQ (Wavelet Scalar Quantization)

have been developed by researchers.

Related Entries

▶ Biometrics, Overview

▶ Biometric Recognition

▶ Fingerprint Anatomy

▶ Fingerprint Classiﬁcation

▶ Fingerprint Compression

▶ Fingerprint Databases and Evaluation

▶ Fingerprint Fake detection

▶ Fingerprint Features

▶ Fingerprint Image Enhancement

▶ Fingerprint Indexing

▶ Fingerprint Individuality

▶ Fingerprint Matching, Automatic

▶ Fingerprint Matching, Manual

▶ Fingerprint Image Quality

▶ Fingerprint Templates

▶ Fingerprints Hashing

References

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

ﬁngerprint recognition. Springer, New York (2003)

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

of identical twin ﬁngerprints. Pattern Recognit. 35(11),

2653–2663 (2002)

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

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

1010–1025 (2002)

4. The Fourth International Fingerprint Veriﬁcation Competition

(FVC2006) http://bias.csr.unibo.it/fvc2006.

5. Lee, H.C., Gaensslen, R.E.: Advances in ﬁngerprint technology.

2nd Edn. Elsevier, New York (2001).

6. Cappelli, R., Lumini, A., Maio, D., Maltoni, D.: Fingerprint

image reconstruction from standard templates. IEEE Trans.

Pattern Anal. Mach. Intell. 29(9), 1489–1503 (2007)

Fingerprint Representation

▶ Fingerprint Templates

Fingerprint Retrieval

Fingerprint retrieval is a procedure that draws a subset

of ﬁngerprints from a database stored on a computer

system based on some similarity measure between a

query ﬁngerprint and the ﬁngerprints in the database.

The ultimate goal of ﬁngerprint retrieval is not to ﬁnd

a group of ﬁngerprints similar to the query ﬁngerprint,

but to get back the ﬁngerprint originating from the

same ﬁnger as that of the query ﬁngerprint. Hence,

success or failure of the ﬁngerprint retrieval is deter-

mined by whether the retrieved subset contains the

ﬁngerprint originating from the same ﬁnger as that of

the query ﬁngerprint.

▶ Fingerprint Classification

▶ Fingerprint Indexing

Fingerprint Sample Synthesis

RAFFAELE CAPPELLI

Biometric System Laboratory, DEIS, University of

Bologna, Cesena, Italy

Synonyms

Synthetic ﬁngerprint generation; Synthetic ﬁnger-

prints; Artiﬁcial ﬁngerprints

Definition

Fingerprint sample synthesis is the generation of

images similar to human ﬁngerprints, through para-

metric models that simulate the main characteristics

of such biometric data and their modes of variation.

The image synthesis is typically performed by a

computer program that, starting from some input

Fingerprint Sample Synthesis

513

parameters, executes a sequence of algorithmic steps

that ﬁnally produce a synthetic ﬁngerprint image.

Introduction

With the increasingly adoption of ﬁngerprint recogni-

tion systems, driven by their very appealing accuracy/

cost tradeoff, methodical and accurate performance

evaluations of ﬁngerprint recognition algorithms are

needed. Unfortunately, this requires large databases

of ﬁngerprints, due to the very small rates of error

necessary for the procedure. For instance, according

to [1], in order to support a claim of FMR less than

1/10,000 (the requirement for veriﬁcation applications

in [2]), 30,000 impostor matches from at least 250

individuals should be performed without observing

any false match error. On the other hand, collecting

large databases of ﬁngerprint images is expensive both

in terms of money and time, boring for both the

people involved and for the volunteers, and problem-

atic due to the privacy legislation that protects such

personal data. FVC competitions [3] are examples of

technolog y evaluations, where real ﬁngerprint data-

bases have been collected to test different algorithms,

but do not constitute lasting solutions for evaluating

and comparing different algorithms; in fact, since FVC

databases are made available to the participants after

the competition to let them improve the technology,

they expire once ‘‘used,’’ and new databases have to be

collected for future evaluations.

Fingerprint synthesis is a feasible way to address the

issues just cited, since it allows large databases of images

to be easily generated and used for testing ﬁngerprint

recognition systems without infringing on privacy.

A ﬁngerprint synthesis method typically consists of

two main steps: ﬁrst, a ridge pattern, which represents

the unique and immutable characteristics of a ‘‘synthetic

ﬁnger,’’ is generated according to a given model; then,

one or more ‘‘ﬁngerprints’’ of the synthetic ﬁnger are

generated by simulating the main factors that make the

ﬁngerprints of a given human ﬁnger different each other.

Physical Ridge Pattern Models

Physical ridge pattern models are based on some

hypothesized physical mechanisms of ﬁngerprint for-

mation during embryogenesis.

The crucial period of ﬁngerprint development in

humans starts at the 10th week of pregnancy [4], when

the epidermis consists of three layers (outside layer,

intermediate layer and basal layer). It is then observed

that the basal layer of the epidermis becomes undulat-

ed toward the surface, forming the so-called ‘‘primary

ridges,’’ whose development ends at about the 17th

week of pregnancy: at this stage the geometry of the

epidermal ridge pattern is determined for life and

becomes visible on the skin surface in subsequent weeks.

Several theories for ﬁngerprint pattern formation

have been proposed in the scientiﬁc literature [4],

including cell proliferation phenomena, mechanical

interaction between the extracellular matrix and ﬁbro-

blasts in the dermis, reaction-diffusion models.

In a study by Sherstinsky and Picard [5], a complex

method which employs a dynamic non-linear system

called ‘‘M-lattice,’’ is introduced. The method is based

on the reaction-diffusion model ﬁrst proposed by Tur-

ing in 1952 to explain the formation of animal patterns

such as zebra stripes. Although this work is aimed at

optimally binarizing a ﬁngerprint image, the underly-

ing ridge-line model could be used as a basis for

synthetic generation.

An interesting model was proposed by Ku

cken

[4, 6], based on the following hypotheses:

1. Fingerprint pattern s are created by forces that are

induced by differential growth of the epidermis’

basal layer (as argued by Cummins [7] from the

observed dependency of the pattern class on the

ﬁngertip geometry)

2. Non-uniform growth of the epidermis’ basal layer

results in compressive stress that leads to bu ckling,

creating the primary-ridges [8]

cken considers the basal layer as an elastic sheet

trapped between the neighboring tissues of the inter-

mediate epidermis layer and the dermis (Fig. 1)

and studied the buckling process by means of the von

Karman equations, which descri be the behavior of a

thin curved sheet of elastic material. The analysis of

those equations conﬁrmed that the direction of the

ridges is roughly perpendicular to the direction of

greatest stress; Ku

cken postulated that two factors

mainly contribute to generate the compressive stress

in the basal layer: (1) resistance at the nail furrow and

at the major ﬂexion creases of the ﬁnger (boundary

effects); (2) the regression of the ‘‘volar pads’’ at the

time of ﬁngerprint development. Volar pads are

514

Fingerprint Sample Synthesis

temporary eminences of the skin surface that form

during the 7th week of pregnancy and start to digress

at about the 10th week. From studies of embryos,

monkeys and malformed hands, it has consistently

been observed that highly rounded pads at the ﬁnger-

tips exhibit whorls; less well-developed pads show

loops, where the direction of the loop opening is de-

termined by the asymmetry of the pad; small indistinct

pads give rise to arches.

Computer simulations have shown results consis-

tent with the above observations and hypothesis; Fig. 2

shows how an almost periodic pattern very similar to

human ﬁngerprints can be generated by applying

cken’s model: the three main ﬁngerprint classes

can be simulated and

▶ minutiae are present in regions

where ridge patches with different directions and/or

wavelength meet.

Statistical Ridge Pattern Models

Statistical ridge pattern models aims to reproduce

realistic-looking ﬁngerprints without starting from

embryological hypothesis. Such models are based on

the empirical analysis of real ﬁngerprints, from which

statistical data about the main characteristics of the

patterns are derived and parameterized into appro-

priate equations or synthesis algorithms.

In 1999, Kosz published some interesting

results concerning ﬁngerprint synthesis based on a

Fingerprint Sample Synthesis. Figure 1 The basal layer of epidermis: Ku

cken and Newell [6] assumes that due to

differential growth, a compressive stress act on this layer.

Fingerprint Sample Synthesis. Figure 2 Simulation of three common fingerprint patters (from left to right: whorl, loop,

and arch) using the model proposed in [6].

Fingerprint Sample Synthesis

515

mathematical model of ridge patterns and minutiae

[9]; further details on this technique have been

provided online by Bicz [10] in 2003. According to

this model, a ﬁngerprint can be described by a wave

pattern:

fx; yðÞ¼cos ’ x; yðÞðÞ ð1Þ

where:

’ x; yðÞ¼’

x; yðÞþ’

x; yðÞ ð2Þ

is a function that deﬁnes the phase of the wave struc-

ture as the sum of two parts: ’

, which describes the

global ‘‘shape’’ of the ridge lines, and ’

, which

describes the minutiae. According to the model intro-

duced by bicz [10], ’

can simply generate n minutiae

by adding n spatially-shifted arctangent functions:

’

x; yðÞ¼

i¼1

arctan

y  y

x  x



ð3Þ

where (x

, y

) is the location of minutia i. Figure 3

shows a synthetic pattern generated by using the

above equations.

In 1993, Sherlock and Monro [11] proposed an

orientation model that allows a consistent

▶ orienta-

tion ﬁeld to be computed from the sole knowledge

of the position of ﬁngerprint

▶ singularities (loops

and deltas). In this model, the image is located in the

complex plane and the local ridge orientation is

the phase of the square root of a complex rational

function whose singularities (poles and zeros) are lo-

cated at the same place as the ﬁngerprint singularities

(loops and deltas). Let ls

, i = 1..n

and ds

, i = 1..n

be the coordinates of the loo ps and deltas respecti-

vely. The orientation y at each point z =[x, y]is

calculated as:

y ¼

i¼1

arg z  ds

ðÞ

i¼1

arg z  ls

ðÞ

ð4Þ

where the function arg(c) returns the phase angle of

the complex number c (see Fig. 4).

The Sherlock and Monro model may be exploited

for generating synthetic orientation ﬁelds by ﬁrst ran-

domly choosing a ﬁngerprint class and then randomly

selecting the positions of the singularities, according to

the class-speciﬁc constraints (for instance, in a left loop,

the delta must be on the right side of the loop). Figure 5

shows some examples of orientation ﬁelds generated

by this model.

However, in nature the ridge-line ﬂow cannot

be completely determined by the singularity ty pe

and position. In 1996, Vizcaya and Gerhardt proposed

a variant of the Sherlock and Monroe mode l that

Fingerprint Sample Synthesis. Figure 3 A simple

synthetic pattern generated by equations (1)–(3),

with ’

x; yðÞ¼20  2p 

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

þ y

and fðx

; y

Þg ¼

fð0:2; 0:25Þ; ð0:2; 0:37Þ; ð0:0; 0:2Þ; ð0 :25; 0:3Þ;

ð0:2; 0:43Þg.

Fingerprint Sample Synthesis. Figure 4 Sherlock and

Monro model: each element of the orientation field is

considered as a complex number.

516

Fingerprint Sample Synthesis

introduces more degrees of freedom to cope with

the orientation variability that may characterize orien-

tation ﬁelds with coincident singularities. The orienta-

tion y at each point z is calculated as:

y ¼

i¼1

arg z  ds

ðÞðÞ

i¼1

arg z  ls

ðÞðÞ

ð5Þ

where g

aðÞ, for k 2 ls

; :::; ls

; ds

; :::; ds

, are

piecewise linear functions capable of locally correcting

the orientation ﬁeld with respect to the value given by

Sherlock and Monroe model:

aðÞ¼



ðÞþ

a  a

2p=L



iþ1

ðÞ



ðÞðÞð6Þ

for a

 a  a

iþ1

, a

¼p þ

2pi

Each function g

aðÞis deﬁned by the set of value s



ðÞi ¼ 0:: L  1

jfg

, where each value is the amount

of correction of the orientation ﬁeld at a given angle

(in a set of L angles uniformly distributed between p

and p). If



ðÞ¼a

8i 2 0::L  1

(i.e. g

aðÞis the

identity function), the model coincides with that of

Sherlock and Monro.

Figure 6a and b show two examples of orientation

ﬁelds generated according to the Vizcaya and Gerhardt

model; these images are deﬁnitely more realistic than

those in Fig. 5. The superiority of the Vizcaya and

Gerhardt model in approximating existing ridge pat-

terns is also evident from the comparison between

Fig. 6c and d.

In 2000, Cappelli et al. introduced a ridge pattern

generation approach based on the following steps [12]:

1. Orientation ﬁeld generation

2. Frequency map generation

3. Ridge pattern generation

Step 1 adopts the Vizcaya and Gerhardt model for

generating the orientation ﬁeld starting from the posi-

tions of loops and deltas; for generating arch type pat-

terns (which do not contain any singularity), a simple

sinusoidal function, whose frequency and amplitude are

tuned to control the arch curvature and aspect, is used.

Step 2 creates a frequency map (see Fingerprint

Feature Extraction) on the basis of some heuristic

criteria inferred by the visual inspection of a large

number of real ﬁngerprints (for instance, in the regions

above the northernmost loop and below the southern-

most delta, the ridge-line frequency is often lower than

in the rest of the ﬁngerprint, see Fig. 7).

Finally step 3, given an orientation ﬁeld and a frequen-

cy map as input, generates a ridge line pattern by itera-

tively enhancing an initial image (containing one or more

isolated points) through

▶ Gabor ﬁlters. The ﬁlters are

applied at each pixel (x, y) and adjusted according to

the local ridge orientation f

and frequency v

gabor r; s : f

; n



¼ e



rþsðÞ

 cos 2pn

r sin f

þ s cos f

hi

ð7Þ

Fingerprint Sample Synthesis. Figure 5 An example of Arch (a), Tented Arch (b), Right Loop (c) and Whorl

(d) orientation ﬁeld as generated by the Sherlock and Monro model. In (e), an example of left-loop orientation ﬁeld

superimposed to a real left-loop ﬁngerprint with coincident singularity positions.

Fingerprint Sample Synthesis

517

Parameter s, which determines the bandwidth of

the ﬁlter, is set according to the frequency, so that the

ﬁlter does not contain more than three effective peaks

(see Fig. 8).

While one could reasonably expect that iteratively

applying ‘‘striped’’ ﬁlters to random images would

simply produce striped images, very realistic minutiae

are generated at random positions. Based on their

experiments, in [12] the authors argue that minutiae

primarily originate from the ridge-line disparity

produced by local convergence/divergence of the

orientation ﬁeld and by frequency changes. In Fig. 9,

examples of the iterative ridge-line generation process

are shown; the authors experimentally found that

increasing the number of initial points determines

a more irregular ridge pattern richer of minutiae: this is

not surprising, since expanding distinct image regions

causes interference where regions merge, thus favoring

the creation of minutiae (see Fig. 10).

Generation of Synthetic Fingerprint

Impressions

Several factors contribute in making the impressions

of a real ﬁnger substantially different when captured

Fingerprint Sample Synthesis. Figure 7 An example of a

right-loop fingerprint where the ridge-line frequency is

lower in the regions above the loop and below the delta.

Fingerprint Sample Synthesis. Figure 6 An example of Right Loop (a) and Whorl (b) orientation ﬁelds, as generated

by the Vizcaya and Gerhardt model. In (c) and (d) the orientation ﬁelds produced by the two models, for a given

ﬁngerprint, are compared.

Fingerprint Sample Synthesis. Figure 8 An example of

Gabor filter used in step 3: note that the bandwidth is

adjusted so that the filter does not contain more than

three peaks.

518

Fingerprint Sample Synthesis

by an on-line acquisition sensor (see ▶ Fingerprint

Device):

1. Displacement in x and y direction and rotation

2. Different touching areas

3. Non-linear distortions produced by non-orthogonal

pressure of the ﬁnger against the sensor

4. Variations in the ridge-line thickness given by pres-

sure intensity or by skin dampness

5. Small cuts or abrasions on the ﬁngertip

6. Background noise and other random noise

In 2002, Cappelli et al. proposed an evolution of the

approach introduced in [13], which is able to simulate

most of the above factors, thus generating very realistic

ﬁngerprint impressions. Starting from a synthetic

ridge-line pattern, the main steps involved in the sim-

ulation of a ﬁngerprint impression are: (1) Variation of

the ridge thickness; (2) Skin distortion; (3) Noising

and global translation/rotation; (4) Background

generation. The subsections that follow brieﬂy

describe the various steps, as they were proposed by

Cappelli [14].

Variation of the Ridge Thickness

Skin dampness and ﬁnger pressure against the sensor

platen have similar effects on the acquired images:

when the skin is dry or the pressure is low, ridges

appear thinner, whereas, when the skin is wet or the

pressure is high, ridges appear thicker (see Fig. 11 ).

Morphological operators (see Image Pr epr oc essing)

are applied to the ridge line pattern, to simulate different

degrees of dampness/pressur e. In particular, the erosion

operator is applied to simulate low pressure or dry

skin, while the dilation operator is adopted to simulate

high pressure or wet skin (see Fig. 12).

Skin Distortion

One of the main aspects that distinguish the different

impressions of the same ﬁnger is the presence of non-

linear distortions, mainly due to skin deformations

Fingerprint Sample Synthesis. Figure 9 Some intermediate steps of a fingerprint-generation process starting from a

single central point (top) and from a number of randomly located points (bottom). Usually, increasing the number of

initial points determines a more irregular ridge pattern richer of minutiae.

Fingerprint Sample Synthesis. Figure 10 Genesis of a minutia point during the merging of the two regions

originated by two different initial points.

Fingerprint Sample Synthesis

519