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

Подождите немного. Документ загружается.

Fingerprint Compression. Figure 2 Comparative example of compressed image of 64  64 pixel region of latent.

Fingerprint Compression. Figure 3 Schematics of decomposition trees, (a) WSQ and (b) JPEG2000 (default).

450

Fingerprint Compression

lengthened middle interval. For the last coding step,

WSQ employs Huffman entropy coding while JPEG2000

uses scalar arithmetic or trellis-coded quantization.

Much of the ﬂexibility and power of the JPEG2000

comes from the Embedded Block Coding with Opti-

mized Truncation (EBCOT) algorithm. Wavelet coefﬁ-

cients from small blocks of the image are processed using

the EBCOT algorithm which adapts the quantizer based

on the statistics of the source image at the bit- plane

level. These blocks are tiled with integrated header infor-

mation concerning with coding details, and further

quantization can be performed on the ﬁnal bit-stream.

The ﬂexibility of this ﬁnal stage in JPEG2000 encoding

greatly assists in providing more optimal compression.

WSQ employs a much simpler approach and coding

details is calculated for each image and the coefﬁcients

included in the ﬁle header.

Fingerprint Compression Performance

WSQ has proved to be satisfactory at compressing

ﬁngerprint images by factors up to about 20, though

the original requirement was to compress to 0.75 bpp

(i.e., 10.7:1 compression). Watson and Wilson [10 ]

report that experiments using WSQ compressed images

with three different matching systems under the con-

ditions that a FAR is 0.001 is maintained and the

maximum reduction in the TAR of less than 0.01 is

permitted, then there is little effect on performance for

compression ratios less than 20:1. A few studies have

compared the relative merits of WSQ and JPEG2000.

Figueroa-Villanueva et al. [11] showed a signiﬁcant

improvement for JPEG2000 over WSQ at 0.75 bpp

compression in terms of PSNR and Receiver Operating

Curves (ROCs) for different sources namely, capaciti ve

sensor, optical sensor, and scanned inked prints.

A study of JPEG2000 and WSQ interoperability [12]

concluded that JPEG2000 produced a slightly lower

quality reconstructed image compared to WSQ for

the same ﬁle size. Most studies have focused on coding

high-quality inked prints or live print capture from

various sources, and not on poorer quality latents. One

study that involved latents and performance on an

operational AFIS system [13] concluded that for com-

pression ratios less than 32:1, JPEG2000 consistently

produced higher identiﬁcation rates than WSQ. There

was also strong indication that moderate degrees of

compression facilitated improved identiﬁcation rates

under normal operating procedures than uncom-

pressed latent images.

Current national and international standards

[14, 15] recommend that WSQ encoding is used

for 500 ppi ﬁngerprint images with compression lim-

ited to 15:1; but for images with resolutions greater

than 500 ppi, 15:1 JPEG2000 should be employed.

The UK national ﬁngerprint system permits latents

to be transmitted and submitted to Ident 1 at 15:1

JPEG2000.

Conclusions

It may appear that satisfactory standards exist for com-

pressing ﬁngerprint images, certainly for the normal

operational requirements associated with the effective

transmission and storage of reference and livescan

images. General enhancements in providing lower-

cost, higher-capacity mass storage and increased band-

width across both ﬁxed and wireless data networks will

reduce the pressure to develop new compression stan-

dards. However, ﬁngerprints are an unusual and fairly

well-deﬁned class of image. Compression schemes such

as JPEG2000 have been developed to cope well for a

very wide variety of imagery and WSQ, though based

on a principled consideration of the statistical proper-

ties of ﬁngerprints, was developed prior to many more

general advances in image compression. Recent propo-

sals for improved ﬁngerprint image compression are

generally based on wavelet transformations but with

more effective decomposition trees, optimized ﬁlter

structures and coefﬁcients through the use of genetic

algorithm optimization, and vector quantization.

The relationship between image enhancement and com-

pression is not fully understood. There are sugges-

tions that the ﬁltering that occurs during compression

may be advantageous in increasing identiﬁcation, espe-

cially for latents. This is an avenue that needs to be

explored further.

Related Entries

▶ Fingerprint Image Enhancement

▶ Fingerprint Matching, Automatic

▶ Fingerprint Matching, Manual

▶ Fingerprint Recognition, Overv iew

Fingerprint Compression

451

References

1. Komarinski, P.: Automated Fingerprint Identiﬁcation Systems

(AFIS). Elsevier Science Technology (2005)

2. Scientiﬁc Working Group on Friction Ridge Analysis, Study and

Technology (SWGFAST): Friction Ridge Digital Imaging Guide-

lines, 1 edn. (2001). URL http://www.theiai.org/guidelines/

swgfast/

3. Ghanbari, M.: Image compression to advanced video coding. In:

Standard Codecs, Telecomunications Series 49. Institution Elec-

trical Engineers (2003)

4. Burrus, C., Guo, H., Gopinath, R.: Introduction to Wavelets and

Wavelet Transforms: A Primer. Prentice Hall (1997)

5. Vetterli, M., Kovacevic, J.: Wavelets and subband coding. Prentice-

Hall,Inc.(1995)

6. Nill, N.: Image quality evaluation - image quality of ﬁngerprint

(IQF). Tech. rep., The MITRE Corporation (2007)

7. Hopper, T., Preston, F.: Compression of grey-scale ﬁngerprint

images. In: Proceedings DCC ’92. Data Compression Confer-

ence, pp. 309–318 (1992). DOI 10.1109/DCC.1992.227450

8. Skodras, A., Christopoulos, C., Ebrahimi, T.: The JPEG 2000 still

image compression standard. IEEE Signal Process Mag 18(5),

36–58 (2001)

9. Antonini, M., Barlaud, M., Mathieu, P., Daubechies, I.: Image

coding using wavelet transform. IEEE Trans Image Process 1(2),

205–220 (1992)

10. Watson, C., Wilson, C.: Effect of image size and compression on

one-to-one ﬁngerprint matching. Tech. rep., National Institute

of Standards and Technology (2005)

11. Figueroa-Villanueva, M., Ratha, N., Bolle, R.: A comparative

performance analysis of JPEG 2000 vs. WSQ for ﬁngerprint

image compression. In: Audio- and Video-Based Biometric

Person Authentication, Lecture Notes in Computer Science,

p. 1059. Springer Berlin/Heidelberg (2003)

12. Lepley, M.: JPEG 2000 and WSQ image compression interoper-

ability. Tech. rep., The MITRE Corporation (2001)

13. Allinson, N.M., Sivarajah, J., Gledhill, I., Carling, M., Allinson, L.J.:

Robust wireless transmission of compressed latent ﬁngerprint

images. **IEEE Transactions on Information Forensics and Secu-

rity 2(3), 331–340 (2007). DOI 10.1109/TIFS.2007.902684

14. American National Standards Institute: American national stan-

dard for information systems - data format for the interchange

of ﬁngerprint, facial, & other biometric information. URL

http://ﬁngerprint.nist.gov/standard/

15. International Organization for Standardization: Information

technology - biometric data interchange formats - part 4: Finger

image data. URL http://www.iso.org/

Fingerprint Contrast Enhancement

▶ Fingerprint Image Enhancement

Fingerprint Corpora

▶ Fingerprint Databases and Evaluation

Fingerprint Data Interchange

Format

▶ Finger Data Interchange Format, Standardization

Fingerprint Databases and

Evaluation

FERNAN DO ALON S O -FERNANDEZ,JULIAN FIERREZ

Biometric Recognition Group – ATVS,

Escuela Politecnica Superior,

Universidad Autonoma de Madrid,

Campus de Cantoblanco, Madrid 28049, Spain

Synonyms

Fingerprint bench mark; Fingerprint corpora

Definition

Fingerprint databases are structured collections of ﬁn-

gerprint data mainly used for either evaluation or

operational recognition purposes.

The ﬁngerprints in databases for evaluation are

usually detached from the identity of the corres-

ponding individuals, are publicly available for research

purposes, and usually consist of raw ﬁngerprint images

acquired with live-scan sensors or digitized from inked

ﬁngerprint impressions on paper. These databases are

the basis for research in a utomatic ﬁngerprint recogni-

tion, and together with speciﬁc experimen tal pro-

tocols, are the basis for a number of technology

evaluations and benchmarks. This is the type of ﬁnger-

print databases further developed here.

452

Fingerprint Contrast Enhancement

On the other hand, ﬁngerprint databases for opera-

tional recognition are typically proprietary, usually in-

corporate personal information about the enrolled

people together with the ﬁngerprint data, and can incor-

porate either raw ﬁngerprint image data or some form

of distinctive ﬁngerprint descriptors such as minutiae

templates. These ﬁngerprint databases repres ent one

of the modules in operational automated ﬁngerprint

recognition systems, and will not be addressed here.

Fingerprint Databases for Evaluation

Among all biometric techniques, ﬁngerprint recogni-

tion is the most widespread in personal identiﬁcation

due to its permanence and uniqueness [ 1]. Finger-

prints are being increasingly used not only in forensic

investigations, but also in a large number of conve-

nience applications, such as access control or online

identiﬁcation [2].

The growth that the ﬁeld has experienced over the

past two decades has led to the appearance of increas-

ing numbers of biometric databases for research and

evaluation purposes, either

▶ monomodal (one bio-

metric trait sensed) or

▶ multimodal (two or more

biometric traits sensed). Previous to the databases ac-

quired within the framework of the International Fin-

gerprint Veriﬁcation Competition series, the only

large, publicly available datasets were the NIST data-

bases [3]. However, these databases were not well

suited for the evaluation of algorithms operating with

live-scan images [1] and will not be described here.

In this section, the authors present some of the most

popular publicly available biometric databases, either

monomodal or multimodal, that include the ﬁnger-

print trait acquired with

▶ live-scan sensors.

FVC Databases

Four international Fingerprint Veriﬁcation Competi-

tions (FVC) have been organized in 2000, 2002, 2004

and 2006 [4, 5, 6, 7]. For each competition, four

databases were acquired using three different sensors

and the SFinGE synthetic generator [1]. Each database

has 110 ﬁngers (150 in FVC2006) with eight impres-

sions per ﬁnger (12 in FVC2006), resulting in 880

impressions (1,800 in FVC2006). In the four competi-

tions, the SFinGe synthetic generator was tuned to

simulate the main perturbations introduced in the

acquisition of the three real databases.

1. In FVC2000 [4], the acquisition conditions were

different for each database (e.g., interleaving/not

interleaving the acquisition of different ﬁngers, peri-

odical cleaning/no cleaning of the sensor). For all

the databases, no care was taken to assure a mini-

mum quality of the ﬁngerprints; in addition, a

maximum rotation and a non-null overlapping area

were assured for impressions from the same ﬁnger.

2. In FVC2002 [5], the acquisition conditions were

the same for each database: interleaved acquisition

of different ﬁngers to maximize differences in ﬁn-

ger placement, no care was taken in assuring a

minimum qualit y of the ﬁngerprints and the sen-

sors were not periodically cleaned. During some

sessions, individual s were asked to: (1) exaggerate

displacement or rotation or, (2) have their ﬁngers

dried or moistened.

3. The FVC2004 databases [6] were collected with the

aim of creating a more difﬁcult benchmark be-

cause, in FVC2002, top algorithms achieved accura-

cies close to 100% [6]. Therefore, more intra-class

variation was introduced. During the different ses-

sions, individuals were asked to: (1) put the ﬁnger

at slightly different vertical position, (2) apply low

or high pressure against the sensor, (3) exaggerate

skin distor tion and rotation, and (4) have their

ﬁngers dried or moistened. No care was taken to

assure a minimum quality of the ﬁngerprints and

the sensors were not periodically cleaned. Also, the

acquisition of different ﬁngers were interleaved to

maximize differences in ﬁnger placement. Effects of

quality degradation in ﬁngerprint images can be

observed in Fig. 1.

4. For the 2006 edition [7], no deliberate difﬁculties

were introduced in the acquisition as it was done in

the previous editions (such as exaggerated distor-

tion, large amounts of rotation and displacement,

wet/dry impressions, etc.), but the population was

more heterogeneous, including manual workers and

elderly people. Also, no constraints were enforced

to guarantee a minimum quality in the acquired

images and the ﬁnal datasets were selected from a

larger database (the BioSec multimodal database

[8]) by choosing the most difﬁcult ﬁngers according

to a quality index, to make the benchmark sufﬁ-

ciently difﬁcult for an evaluation.

Fingerprint Databases and Evaluation

453

BIOMET Multimodal Database

Five different biometric modalities are present in the

BIOMET database [9]: audio, face image, hand image,

ﬁngerprint and signature. This database was designed

with the additional goal of incl uding unusual sensors

(face images captured with an infrared camera and

with a 3D acquisition system). The database consists

of three different acquisition sessions. The number of

individuals participating to the collection of the data-

base was 130 for the ﬁrst session, 106 for the second,

and 91 for the last one, resulting in 91 individuals who

completed the whole acquisition process. For ﬁnger-

print acquisition, an optical and a capacitive sensor

were used. During the ﬁrst acquisition campaign, only

the optical sensor was used, whereas both the optical

and capacitive sensors were employed for the second

ant third campaigns. The total number of available

ﬁngerprints per sensor in the BIOMET database is

6 for the middle and index ﬁngers of each contributor.

MCYT Bimodal Database

A large biometric database acquisition process was

launched in 2001 by four Spanish academic institu-

tions within the MCYT project [10]. The MCYT data-

base includes ten-print ac quisition (MCYT Fingerp rint

subcorpus) and on-line signature (MCYT Signature sub-

corpus) samples of each individual enrolled in the data-

base. A total of 330 individuals were acquired in the four

institutions participating in the MCYT project. R eg ard-

ing the MCYT Fingerprint subcorpus, for each individ-

ual, 12 samples of each ﬁnger were acquired using

an optical and a capacitive sensor under differ ent con-

trol conditions. The MCYT database has been extended

with the comprehensive BiosecurID multimodal data-

base [11], which includes 8 different biometric traits

from 400 donors collected in 4 sessions separated in

time.

BioSec Multimodal Database

BioSec was an Integrated Project of the Sixth European

Framework Programme which involved over 20 part-

ners from nine European countries. The goal of BioSec

was to leverage the integration of biometrics in a wide

spectrum of everyday’s applications. One of the activ-

ities within BioSec was the acquisition of a multimodal

database. This database was acquired at four different

European sites and includes face, speech, ﬁngerprint

and iris recordings. The baseline corpus [8] comprises

200 subjects with two acquisition sessions per subject.

The extended version of the BioSec database comprises

250 subjects with four sessions per subject (about

1 month between sessions). Each subject provided in

each session four samples of each of four ﬁngers

(left and right index and middle). Fingerprints were

acquired using three different sensors. Some example

images are shown in Fig. 2.

BioSecure Multimodal Database

The acquisition of the BioSecure Multimodal Database

(BMDB) was jointly conducted by 11 European insti-

tutions participating in the BioSecure Network of

Excellence. [11] The BMDB is comprised of three

different datasets [12], namely:

1. Data Set 1 (DS1 ) , acquired over the Internet under

unsupervised conditions (i.e., connecting to an

Fingerprint Databases and Evaluation. Figure 1 Examples of quality degradation in fingerprint images due to factors

like low/high pressure, dryness/moisture, dirt, etc.

454

Fingerprint Databases and Evaluation

URL and following the instructions provided on

the screen).

2. Data Set 2 (DS2), acquired in a standard ofﬁce

room environment using a PC and a number of

commercial sensors under the guidance of a human

supervisor.

3. Data Set 3 (DS3), acquired using two mobile

▶ hand-held devices under two acquisition con-

ditions (controlled-indoor and uncontrolled-

outdoor).

The three datasets of the BMDB include a common

part of audio and v ideo data. Additionally, DS2

includes signature, ﬁngerprint, hand and iris data,

and DS3 includes signature and ﬁngerprint data. The

three datasets were acquired in two different sessions

(approximately 2 months between them). Pending yet

to be distributed publicly, the BioSecure multimodal

database has approximately 1,000 subjects in DS1,

and 700 in DS2 and DS3. Fingerprint data in DS2

were acquired using an optical and a capacitive sensor.

Fingerprint data in DS3 were acquired with a PDA.

The databases MCYT, BiosecurID, BioSec, and Bio-

Secure have some commonalities that enable their

integration for speciﬁc research studies, e.g., on time

variability and sensor interoperability [12].

Fingerprint Evaluation Campaigns

The most important evaluation campaigns carried out

in the ﬁngerprint modality are the NIST Fingerprint

Vendor Technology Evaluation (FpVTE2003) [13] and

the four Fingerprint Veriﬁcation Competitions (FVC),

which took place in 2000 [4], 2002 [5], 2004 [6] and

2006 [7]. A comparative summary between FVC2004,

FVC2006 and FpVTE2003 is given Table 1. An impor-

tant evaluation is also the NIST Minutiae Interopera-

bility Exchange Test (MINEX) [14].

Fingerprint Verification

Competitions (FVC)

The Fingerprint Veriﬁcation Competitions were orga-

nized with the aim of determining the state of the art

in ﬁngerprint veriﬁcation. These competitions have

received great attention both from academic and com-

mercial organizations, and several research groups

have used the FVC datasets for their own experiments

later on. The number of participants and algorithms

evaluated has increased in each new edition of the

FVC. Also, to increase the number of participants,

anonymous participation was allowed in 2002, 2004

and 2006. Additionally, the FVC2004 and FVC2006

were subdivided into: (1) open category and (2) light

category. The light category aimed at evaluating algo-

rithms under low computational resources, limited

memory usage and small template size.

For each FVC competition, four databases were

acquired using three different sensors and the SFinGE

synthetic generator [1]. The size of each database

was set at 110 ﬁngers with eight impressions per

ﬁnger (150 ﬁngers with 12 impressions per ﬁnger in

FVC2006). A subset of each database (all the impres-

sions from ten ﬁngers) was made available to the

participants prior to the competition for algorithm

tuning. The impressions from the remaining ﬁngers

were used for testing. Once tuned, participants

submitted their algorithms as executable ﬁles to the

evaluators. The executable ﬁles were then tested at the

evaluator’s site and the test data were not released until

Fingerprint Databases and Evaluation. Figure 2 Example fingerprint images of two fingers acquired with

three different sensors (from the BioSec baseline corpus). Fingerprint images of the same finger are shown for

a capacitive sensor (left of each subplot), an optical sensor (center) and a thermal sensor (right).

Fingerprint Databases and Evaluation

455

the evaluation concluded. In order to benchmark the

algorithms, the evaluation was divided into: (1)

▶ gen-

uine attempts: each ﬁngerprint image is compared to

the remaining images of the same ﬁnger, and (2)

▶ im-

postor attempts: the ﬁrst impression of each ﬁng er is

compared to the ﬁrst image of the remaining ﬁngers.

In both cases, symmetric matches were avoided.

In Table 2, results of the best performing algorithm

in each FVC competition are shown. Data in the 2000

and 2002 editions were acquired without special

restrictions and, as observed in Table 2, error rates

decrease signiﬁcantly from 2000 to 2002, demonstrating

in some sense the maturity of ﬁngerprint veriﬁcation

systems. However, in the 2004 and 2006 editions, it is

observed that error rates increase with respect to the

2002 edition due to the deliberate difﬁculties introduced

in the data, thus revealing that degradation of quality

has a severe impact on the recognition rates [15].

Fingerprint Databases and Evaluation. Table 2 Results in terms of equal error rate (EER) of the best performing

algorithm in each of the four databases of the FVC competitions

Database 2000 2002 2004 2006

DB1 (%) 0.67 0.10 1.97 5.56

DB2 (%) 0.61 0.14 1.58 0.02

DB3 (%) 3.64 0.37 1.18 1.53

DB4 (%) 1.99 0.10 0.61 0.27

Average 1.73 0.19 2.07 2.16

Fingerprint Databases and Evaluation. Table 1 Comparative summary between FVC2004, FVC2006 and FpVTE2003

(adapted from [6])

Evaluation FVC 2004 FVC 2006 FpVTE 2003

Algorithms Open category: 41

Light category: 26

Open category: 44

Light category: 26

Large scale test (LST): 13

Medium scale test (MST): 18

Small scale test (SST): 3

Population Students Heterogeneous (including

manual workers and elderly

people)

Operational data from a variety of

U.S. Government sources

Fingerprint format Flat impressions from

low-cost scanners

Flat impressions from

low-cost scanners

Mixed formats (flat, slap and rolled)

from various sources (paper cards,

scanners)

Perturbations Deliberately exaggerated

perturbations

Selection of the most difficult

images according to a quality

index

Intrinsic low quality fingers and/or

non-cooperative users

Data collection Acquired for this event From the BioSec database From existing U.S. Government

sources

Database size Four databases, each

containing 880 fingerprints

from 110 fingers

Four databases, each

containing 1,800 fingerprints

from 150 fingers

48,105 fingerprints from 25,309

subjects

Anonymous

participation

Allowed Allowed Not allowed

Best average EER

(over all the

databases used)

2.07 % (Open category) 2.16 % (Open category) 0.2 % (MST, the closest to the FVC

open category)

456

Fingerprint Databases and Evaluation

NIST Fingerprint Vendor Technology

Evaluation (FpVTE2003)

The NIST Fingerprint Vendor Technology Evaluation

(FpVTE2003) [13] aimed at: (1) comparing systems on

a variety of ﬁng erprint data and identifying the most

accurate systems; (2) measuring the accuracy of ﬁnger-

print matching, identiﬁcation, and veriﬁcation on ac-

tual operational ﬁngerprint data; and (3) determining

the effect of a variety of variables on matcher accuracy.

Eighteen different companies competed in the FpVTE,

and 34 systems were evalua ted.

Three separate subtests were performed in the

FpVTE2003: (1) the large-scale test (LST), (2) the

medium-scale test (MST), and (3) the small-scale test

(SST). SST and MST tested matching accuracy using

individual ﬁngerprints, whereas LST used sets of ﬁn-

gerprint images. The size and structure of each test

were designed to optimize competing analysis objec-

tives, available data, available resources, computational

characteristics of the algorithms and the desire to in-

clude all qualiﬁed participants. In particular, the sizes

of MST and LST were only determined after a great deal

of analysis of a variety of issues. Designing a well-

balanced test to accommodate heterogeneous system

architectures was a signiﬁcant challenge.

Data in the FpVTE2003 came from a variety of U.-

S. Government sources, including low quality ﬁngers of

low quality sources. 48,105 sets of ﬂat slap or rolled

ﬁngerprint sets from 25,309 individuals were used,

with a total of 393,370 ﬁngerprint images. The systems

that resulted in the best accuracy performed consistently

well over a variety of image types and data sources. Also,

the accuracy of these systems was considerably better

than the rest of the systems. Further important conclu-

sions drawn from the FpVTE2003 included: (1) the

number of ﬁngers used and the ﬁngerprint quality had

the largest effect on system accuracy; (2) accuracy on

controlled data was signiﬁcantly higher than accuracy on

operational data; (3) some systems were highly sensitive

to the sources or types of ﬁngerprints; and (4) accuracy

dropped as subject age at time of capture increased.

NIST Minutiae Interoperability Exchange

Test (MINEX)

The purpose of the NIST Minutiae Inte roperability

Exchange Test (MINEX) [14] was to determine the

feasibility of using minutiae data (rather than image

data) as the interchange medium for ﬁngerprint infor-

mation between different ﬁngerprint matching sys-

tems, and to quantify the veriﬁcation accuracy

changes when minutiae from dissimilar systems are

used for matching ﬁngerprints.

▶ Interoperability of

templates is affected by the method used to encode

minutiae and the matcher used to compare the tem-

plates. There are different schemes for deﬁning the

method of locating, extr acting, formatting and match-

ing the minutiae information from a ﬁngerprint image

[1]. In the MINEX evaluation, proprietary template

formats were compared to the ANSI INCITS 378-

2004 template standard.

The images used for this test came from a variety of

sensors, and included both live-scanned and non live-

scanned rolled and plain impression types. No latent

ﬁngerprint images were used. Participants submitting

a system had to provide an algorithm capable of

extracting and matching a minutiae template using

both their proprietary minutiae format and the ANSI

INCITS 378-2004 minutiae data format standard. The

most relevant results of the MINEX evaluation are:

1. In general, proprietary templates lead to better

recognition performance than the ANSI INCITS

378-2004 template.

2. Some template generators produce standard tem-

plates that are matched more accurately than others.

Some matchers compare templates more accurately

than others. The leading vendors in generation are

not always the leaders in matching and vice-versa.

3. Authentication accuracy of some matchers can be

improved by replacing the vendors template gener-

ator with that from another vendor.

4. Performance is sensitive to the quality of the data-

set. This applies to both proprietary and interoper-

able templates. Higher quality datasets provide

reasonable interoperability, whereas lower quality

datasets do not.

Related Entries

▶ Biometric Sample Acquisition

▶ Fingerprint Device

▶ Interoperability

▶ Performance

▶ Performance Evaluation

Fingerprint Databases and Evaluation

457

References

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

Fingerprint Recognition. Springer, New York (2003)

2. Jain, A., Ross, A., Pankanti, S.: Biometrics: A tool for informa-

tion security. IEEE Trans. Inf. Forensics Secur. 1, 125–143 (2002)

3. NIST Special Databases and Software from the Image Group,

http://www.itl.nist.gov/iad/894.03/databases/defs/dbases.html

4. Maio, D., Maltoni, D., Capelli, R., Wayman, J., Jain, A.:

FVC2000: Fingerprint veriﬁcation competition. IEEE Trans. Pat-

tern Anal. Mach. Intell. 24, 402–412 (2002)

5. Maio, D., Maltoni, D., Capelli, R., Wayman, J., Jain, A.:

FVC2002: Second ﬁngerprint veriﬁcation competition. Proc.

Intl. Conf. Pattern Recognit. 3, 811–814 (2002)

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

Performance evaluation of ﬁngerprint veriﬁcation systems. IEEE

Trans. Pattern Anal. Mach. Intell. 28, 3–18 (2006)

7. FVC2006. Fingerprint Veriﬁcation Competition, http://bias.csr.

unibo.it/fvc2006/default.asp (2006)

8. Fierrez, J., Ortega-Garcia, J., Torre-Toledano, D., Gonzalez-

Rodriguez, J.: BioSec baseline corpus: a multimodal biometric

database. Pattern Recognit. 40, 1389–1392 (2007)

9. Garcia-Salicetti, S., Beumier, C., Chollet, G., Dorizzi, B., les

Jardins, J., Lunter, J., Ni, Y., Petrovska-Delacretaz, D.: BIOMET:

A multimodal person authentication database including face,

voice, ﬁngerprint, hand and signature modalities. Lecture

Notes Comput. Sci. 2688, 845–853 (2003)

10. Ortega-Garcia, J., Fierrez-Aguilar, J., Simon, D., Gonzalez, J.,

Faundez-Zanuy, M., Espinosa, V., Satue, A., Hernaez, I., Igarza,

J., Vivaracho, C., Escudero, D., Moro, Q.: MCYT baseline cor-

pus: a bimodal biometric database. IEE Proc. Vision Image

Signal Process 150, 395–401 (2003)

11. Fierrez, J., Galbally, J., Ortega-Garcia, J., et al: BiosecurID:

A multimodal biometric database. Pattern Anal. Appl. 12 (2009)

12. Ortega-Garcia, J., Fierrez, J., Alonso-Fernandez, F., et al: The

multi-scenario multi-environment BioSecure multimodal

database (BMDB), IEEE Trans. Pattern Anal. Mach. Intell.

31 (2009)

13. Wilson, C., et al: Fingerprint Vendor Techonology Evaluation

2003: Summary of Results and Analysis Report. NISTIR 7123,

http://fpvte.nist.gov (2004)

14. Grother, P., et al.: MINEX – Performance and interoperability

of the INCITS 378 ﬁngerprint template, NISTIR 7296, http://

ﬁngerprint.nist.gov/minex (2005)

15. ANSI-INCITS 378, Fingerprint Minutiae Format for Data Inter-

change, American National Standard (2004)

Fingerprint Device

▶ Fingerprint, Palmprint, Handprint and Soleprint

Sensor

Fingerprint Encryption

▶ Fingerprints Hashing

Fingerprint Fake Detection

JEAN-FRANC¸OIS MAINGUET

Grenoble, France

Synonyms

Liveness detection; Cut ﬁnger problem; Dead ﬁnger

detection; Fake ﬁnger detection; Gummy bear ﬁnger;

Latex ﬁnger; Liveness detection

Definition

Fingerprint fake detection is used to identify a fake

ﬁnger, such as a cast made of latex. By extension, it

also includes tests to detect a cut ﬁnger or dead ﬁnger,

or a latent print remaining on a sensor after usage.

Introduction

In ‘‘Diamonds are forever’’ (1971) [1] James Bond took

the identity of Peter Frank with a thin layer of latex

glued on his ﬁngertip to spoof Tiffany Case’s camera.

James was using a simple fake ﬁnger, but the situa-

tion can be worse. With automated ﬁngerprint recogni-

tion systems becoming more widely used, concerns over

ﬁngerprint fake detection have increased. In March

2005, a team of carjackers in Subang Jaya in Malaysia

chopped off part of the owner’s left index ﬁnger, when

they realized that his S-Class Mercedes Benz had a

security feature which would immobilize the car with-

out his ﬁngerprint. Even with more reliable cut ﬁnger

detectors in use, it is likely that this will happen again.

Security of a ﬁngerprint-based system can be

divided into two main areas:

1. The electronic security, which poses the question:

‘‘Is the electronic system, at the other end of the

wires, a real trustful authorized ﬁngerprint system ?’’

458

Fingerprint Device

2. The liveness securit y, which asks a different ques-

tion: ‘‘Is the object touching the sensor a real ﬁnger,

alive and connected to a living person?’’

Answers for electronic security deal with cryptography,

using challenge-response schemes and cryptographic

codes. Since the focus of this essay is to answer the

second question, we will suppose that the electronic

system is perfect and cannot be broken.

To begin, we know 100% security does not exist.

However, what we would verify is that, ‘‘I’m Mr X, a

living person not under threat and I agree to this

action.’’ Lacking the ability to read a person’s mind,

this is an impossible task. At the opposite end, a basic

ﬁngerprint system w ill identify a particular ﬁngerprint

image as likely the same one as registered in the tem-

plate, which is only a small brick within a full security

system.

To ﬁll the gap, we need to acquire more informa-

tion that will enable us to say ‘‘this is a real alive ﬁnger.’’

If we can do that, then we have a good chance to know

that a real person is making the transaction, rather a

cast or cut ﬁnger being applied to the sensor. This will

not answer the problem of detecting a person under

threat, but it should be enough under normal usage,

although some situations will never be detectable. For

example, it will be impossible to detect a graft. In

France, a man received two hands from a donor, a

great medical achievement [2]. But at the same time,

he received 10 brand new ﬁngerprints! There is also

the case of George who attempted to enter the US

illegally on 24 September 2005 through the Nogales,

Arizona Port of Entry during which time US Customs

and Border Protection ofﬁcers noted that his ﬁnger-

prints had been surgically replaced with skin from

his feet. George stated that this procedure had been

done by a doctor in Phoe nix to ‘‘clean’’ his identity [3].

But these should be extremely rare cases. What is

primarily desired is to avoid anyone stealing a ﬁnger-

print to impersonate someone else. So, while it is

impossible to create an absolute fake ﬁnger detection

system, it is possible to make things extremely hard

to be cracked.

Compromised Fingerprint

When someone creates a fake of one of the ﬁngerprints

and use it to spoof a ﬁngerprint system, then we say

that this ﬁngerprint is compromised. With a smart

card (or a key), the smart card can be revoked. Further

use of the card can be prevented and a new one can be

created. But with ﬁngerprints, this is limited to the

10 ﬁngers. Biometric traits – the basis of biometrics

cannot be revoked.

Liveness detection solves the compromised ﬁnger-

print problem. If the system can check that it is the

real alive ﬁnger, then there is no possibility of using

a fake.

Attended/Unattended System

It is commonly admitted that an attended biometric

system does not need any liveness detection because

the superv isor ‘‘obviously’’ checks that a real alive

person is present. In the case of ﬁngerprints, this

would be true if the supervisor was checking the ﬁnger:

is the ﬁnger really connected to the body, and without

any glued cast (Fig. 1)?

Fingerprint Fake Finger Detection Levels

There are three fake ﬁnger methods and detection

levels described, starting from the easiest to the hardest

to detect:

1. Latent print left on the sensor

2. Fake/copies:

a. Printed ﬁngerprint imag e

b. Fake made of gelatin, latex, or other material

Fingerprint Fake Detection. Figure 1 Thin fake made of

gelatine glued on a real finger.

Fingerprint Fake Detection

459