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

Подождите немного. Документ загружается.

voice of the target speaker. In practice, however, cur-

rent state-of-the-art text-to-speech synthesis is not

quite capable of producing such natural sounding

utterances. In other words, synthetic speech produced

by current TTS systems still sounds far from natural

and is easily distinguished from genuine human speech

by the human ear. Does this mean, however, that TTS

speech could not deceive an authentication system

based on automatic speaker recognition? To answer

this question, it needs to be examined how different

speaker recognition systems actually work.

As shown in Table 1, there are three types of speaker

recognition systems that are distinct by the types of

speech patterns each examines in order to determine

the similarity of the unknown speech and the target

speech. The most common type of speaker recognition

system looks at speaker differences at the individual

sound level. A second type of speaker recognition

system examines the sequences of speech sounds,

which form words, and a third type also analyzes

higher-level information such as intonation, choice

of words, choice of sentence structure or even semantic

or pragmatic content of the utterances in question [4].

Speech processing invariably segments a speech

signal into small chunks, or ‘‘frames’’, which corre-

spond in duration roughly to short speech sounds,

say about 10–30 ms. For each frame, features are

extracted from the speech signal, such as a spectrum

or a cepstrum or a mel-frequency cepstrum (MFC) [5].

These extracted features serve as the basis for the

comparison between the unknown speech and the

target speech. The ﬁrst type of speaker recognit-

ion system independently compares the features of

each frame of the unknown speech signal with

the model of the target speaker. This is done indepen-

dently for each frame and without considering the

speech sounds immediately preceding or succeeding

the given frame.

The second type of speaker recognition system

takes into account the likelihood of sequences of speech

sounds, rather than individual speech sounds, when

comparing the unknown speech signal with the model

of the target speaker. For example, the sound sequence

/the/ would be more likely for a speaker of Eng lish than

the sound sequence /eth/. The third type of system

takes into account higher-level features, i.e., the varia-

tion of features over time such as the intonation

pattern of a sentence, as it manifests itself through

the functions of loudness and pitch over time. Such

authentication systems typically operate on much lon-

ger passages of speech, for example, to segment a two-

way telephone conversation or the proceedings in a

court of law into the turns belonging to the different

speakers. Figure 4 shows an example of two speakers

pronouncing the same sentence with quite different

intonation.

It is easy to see that a context-free authenticati on

system is prone to be attacked successfully by a very

simple synthesizer, namely one that produces a few

seconds of only a single speech sound. For example,

an attacker could reproduce a single frame of, say, the

sound ‘‘a’’ of the target speaker and play this frame

Liveness Assurance in Voice Authentication. Table 1 Types of speaker authentication methods

Type of speaker

recognition system Training/Enrolment Testing Typical method

Recognizes

individual speech

sounds (context-

free)

A set of speech sounds typical for

the target speaker is collected and

becomes the ‘‘model’’ for the target

speaker

Each speech sound is individually

compared with the target speaker

‘‘model’’

Gaussian

MixtureModel (GMM)

Recognizes

sequences of

speech sounds

(context-sensitive)

In addition to the individual

sounds, the speaker model

represents the sequences of speech

sounds that are typical for the

target speaker

The entire utterance is compared

with the target speaker model for

both individual sounds and sound

sequences

Hidden Markov Model

(HMM)

Recognizes higher-

level features

(intonation, word

choice, syntax etc.)

In addition to sound sequences, the

speaker model represents words,

sentence structures and intonation

patterns typical for the target

speaker

Similarity of sounds and sound

sequences is combined with

similarity of word sequences and

intonation patterns

Information fusion of

GMM and/or HMM

with higher-level

information sources

920

Liveness Assurance in Voice Authentication

Liveness Assurance in Voice Authentication. Figure 4 Two male speakers from the Australian National Database of

Spoken Language (ANDOSL), speaking the same sentence with distinctly different intonation: audio signal, and

power and fundamental frequency (F0) contours. Speaker S017 produced the word John with falling F0, while Speaker

S029 produced the same with rising F0.

Liveness Assurance in Voice Authentication

921

repeatedly for a second or two in order to ‘‘convince’’

an authentication system of this type that the

‘‘aaaaaaa...’’ sound represents the natural voice of the

target speaker. This is because each frame is assessed

independently as being similar to the ‘‘a’’ sound of the

target speaker, irrespective of the fact that the sequence

of ‘‘a’’ sounds does not represent a likely speech pattern

of the target voice.

A context-sensitive authentication system, on the

other hand, requires a speech synthesizer to reproduce

entire sound sequences that are sufﬁciently similar to

sound sequences produced by the target speaker. This

means that the individual sounds, produced by the

synthesizer, must be similar to sounds of the target

speaker and the sound sequences must be structured

in a similar way to those of the target speaker. This is a

proposition that is far more difﬁcult, although not

impossible, to achieve with current-state-of-the-art

speech synthesizers. Furthermore, if the speaker au-

thentication system also considers the intonation pat-

tern and higher-level features such as choice of words

and grammatical constructs, an attacker who tries to

impersonate a target speaker using a TTS synthesizer,

would require a system that is beyond the capabilities

of the technology at the time of writing.

Multimodal Liveness Assurance

The assurance that a voice biometric is delivered live at

the time and place of authentication can be enhanced

considerably by complementing the voice modality

with a second modality. In the simplest case, this

could be the visual modality provided by a human

observer who can assure that the voice biometric is

actually provided by the person seeking authentication

and that person is not using any device to play back a

recorded or synthesized voice sample.

In an automatic voice authentication system, simi-

lar assurance of liveness can be achieved by combining

the voice modality with a face recognition system. Such

a system has a number of advantages. Firstly, the bi-

modal face-voice approach to authentication provides

two largely independent feature sets, which, when

combined appropriately, can be expected to yield bet-

ter authentication than either of the two modalities by

itself. Secondly, the bimodal approach will add robust-

ness to the system when either mod ality is affected by

difﬁcult environmental conditions. In the case of bi-

modal face-voice authentication, it is par ticularly use-

ful to fall back on the complementary face recognition

facility when the voice recognition modality breaks

down due to high levels of surrounding noise, compet-

ing speakers or channel variability such as that caused

by weak cell phone reception. In such situations, the

face recognition modality will be able to take over and

hence provide enhanced robustness for the combined

system.

A simila r consideration applies, of course, when the

combined face-voice authentication system is viewed

from the perspective of the face recognition modality,

which may equally break down in difﬁcult environ-

mental conditions such as adverse lighting. In this case,

too, the overall robustness of the authentication system

is preserved by the combination of the two modalities

voice and face, each of which is affected differently and

largely independently by environmental factors.

However, the most important advantage of a bi-

modal face-voice authentication system for the assur-

ance of liveness is the fact that the articulator

movements, mainly of the lips, but also of the tip of

the tongue, jaw, and cheeks are mostly observable and

correspond closely to the particular speech sounds

produced. Therefore, it is possible when observing a

bimodal audio-video signal of the speaking face to

ascertain whether the facial dynamics and the sequence

of speech sounds are mutually compatible and

synchronous. To a human observer it is quite discon-

certing when this is not the case, for example, with an

out-of-sync television signal or with a static facial

image when the speaker is heard saying something,

but the lips are not seen to be moving. In the ﬁeld of

audiovisual speech recognition, the term ‘‘viseme’’ has

been coined as the visual counterpart of the ‘‘

▶ pho-

neme’’, which denotes a single speech sound. The

visemes /m/, /u/, and /d/ (as in the word ‘‘mood’’),

for example, ﬁrst show the speaker’s lips spread and

closed (for /m/), then protruded and rounded (for /u/),

and ﬁnally spread and slightly open (for /d/). It is

therefore possible to detect whether the corresponding

sequences of visemes and phonemes of an utterance

are observed in a bimodal audio-video signal and

whether the observed viseme and phoneme sequences

are synchronous.

In order for the synchrony of the audio and video

streams to be ascer tained, the two modalities must be

combined appropriately. Multimodal authentication

922

Liveness Assurance in Voice Authentication

systems employ different paradigms to combine, or

‘‘fuse’’, information from the different modalities. Mo-

dality fusion can happen at different stages of the

authentication process. Fusing the features of the dif-

ferent channels immediately after the feature extrac-

tion phas e is known as ‘‘feature fusion’’ or ‘‘early

fusion’’. In this paradigm, all comparisons between

the unknown sample and the client model as well as

the decision making are based on the combined feature

vectors. The other possibility is to fuse information

from the two modalities after independent compari-

sons have been made for each modality. Such para-

digms are known as

▶ score fusion, ▶ decision fusion

▶ late fusion.

For liveness assurance by means of bimodal face-

voice authentication, it is necessary to apply an early

fusion stratagem, i.e., to fuse the two modalities at the

feature level [6]. If the two modalities were fused late,

i.e., at the score or decision level, analysis of the video of

the speaking face would yield one decision on the speak-

er’s identity and analysis of the audio of the utterance

would yield another decision on the speaker’s identity.

The two processes would run independently of each

other with no connection between them that would

allow the checking for the correspondence and syn-

chrony of visemes and phonemes [7].

Therefore, the features that are extracted from the

audio signal on a frame-by-frame basis – usually at an

audio frame rate of about 40–100 frames per second –

must be combined with the features that are extracte d

from the video signal – usually at the video frame rate

of 25 or 30 frames per second. An example of how the

differing frame rates for the audio and video signals

can be accommodated is shown in Fig. 5, where the

audio frame rate is 50 frames per second, the video frame

rate is 25 frames per second, and the combined audiovi-

sual feature vector comprises the audio feature vectors of

two consecutive audio frames, combined with the single

video vector of the synchronous video frame.

The combined audiovisual feature vectors will then

reveal whether the audio and video streams are synchro-

nous, for example, when the combined audiovisual

feature vectors contain the sequence of visemes /m/,

/u/, and /d/ and likewise the sequence of phonemes

/m/, /u/, and /d/. In contrast, if one of the combined

audiovisual feature vectors were to contain the visual

information for the viseme /m/ and at the same

time the audio information for the phoneme /u/, the

combined feature vector would indicate that the audio

and video streams do not represent a corresponding

synchronous representation of any speech sound.

The proper sequencing of visemes and phonemes

is usually ascertained by representing the audio-

visual speech by Hidden Markov Models (HMM),

which establish the likelihoods of the different com-

bined audiovisual vectors and their sequences over

time [8]. It is therefore possible to ascertain whether

the audio and video components of a combined

audio-video stream represent a likely live utterance.

Therefore, an attacker who attempts to impersonate

a target speaker by means of a recorded speech utter-

ance and a still photograph of the target speaker will

be thwarted because the system will recognize the fail-

ure of the face to form the corresponding visemes that

should be observed synchronously with the phoneme s

of the utterance. Similarly, such a system will thwart an

attack by an audiovisual speech synthesis system, un-

less the synthesizer can generate the synthe tic face and

the synthetic voice in nearly perfect synchrony.

Related Entries

▶ Biometric Vulnerabilities: Overview

▶ Security and Liveness, Overview

▶ Voice Authentication

Liveness Assurance in Voice Authentication. Figure 5

Feature fusion of two consecutive 20ms audio feature

vectors with the corresponding 40ms video feature vector.

Before fusion, the audio vectors have been reduced to

8 dimensions each, and the video vector has been reduced

to 20 dimensions. The combined feature vector has

36 dimensions.

Liveness Assurance in Voice Authentication

923

References

1. Furui, S.: Cepstral analysis techniques for automatic speaker

veriﬁcation. IEEE Trans. Acoust., Speech and Signal Processing

ASSP-29, 254–272 (1981)

2. Bimbot, F., Bonastre, J.-F., Fredouille, C., Gravier, G., Magrin-

Chagnolleau, I., Meignier, S., Merlin, T., Ortega-Garcı

a, J.,

Petrovska-Delacre

taz, D., Reynolds, D.A.: A tutorial on text-

independent speaker veriﬁcation, EURASIP J. Appl. Signal Pro-

cessing 2004(4), 430–451 (2004)

3. Matsui, T., Furui, S.: Speaker adaptation of tied-mixture-based

phoneme models for text-prompted speaker recognition. In:

Proceedings of International Conference on Acoustics, Speech

and Signal Processing , Adelaide, pp. I-125–128. IEEE Publisher,

New York (1994)

4. Reynolds, D., Andrews, W., Campbell, J., Navratil, J., Peskin,

B., Adami, A., Jin, Q., Klusacek, D., Abramson, J., Mihaescu,

R., Godfrey, J., Jones, D., Xiang, B.: The SuperSID Project:

Exploiting high-level information for high-accuracy speaker

recognition. In: Proceedings of International Conference on

Acoustics, Speech and Signal Processing, Hong Kong, pp. IV-

784–787. IEEE Publisher, New York (2003)

5. Huang, X., Acero, A., Hon, H.-W.: Spoken language processing,

Prentice Hall, Upper Saddle River, NJ (2001)

6. Chetty, G., Wagner, M.: Investigating feature-level fusion for

checking liveness in face-voice authentication. In: Proceed-

ings of eighth IEEE Symposium on Signal Processing and

its Applications, Sydney, pp. 66–69. IEEE Publisher, New York

(2005)

7. Bredin, H., Chollet, G.: Audiovisual speech synchrony measure:

Application to biometrics. EURASIP J. Adv. Signal Process. 2007

(1), 1–11 (2007)

8. Chetty, G., Wagner, M.: Speaking faces for face-voice speaker

identity veriﬁcation. In: Proceedings of Interspeech-2006 –

International Conference on Spoken Language Processing,

Paper Mon3A1O-6, Pittsburgh. International Speech Commu-

nication Association (2006)

Liveness Detection

Liveness detection is a functionality that determines

whether the presented biometric sample (e.g., ﬁnger,

hand, or iris) is originated from a live body. This func-

tionality is considered to be one of the key security

measures that improve the reliability of a biometric sys-

tem because it enables the system to reject artifacts to be

enrolled and ensure that no forged sample is accepted.

▶ Finger Vein Reader

Liveness Detection

In biome tric systems, the goal of liveness testing is to

determine if the biometric being captured is an actual

measurement from the authorized, live person who is

present at the time of capture. While biometric systems

may have an excellent performance and improve secu-

rity, previous studies have shown it is not difﬁcult to

spoof biometric devices through fake ﬁngers, high

resolution images or video, contact lenses, etc. Even

though biometric devices use physiologic information

for identiﬁcation/veriﬁcation purposes, these measure-

ments rarely indicate liveness. Liveness detection

reduces the risk of spooﬁng by requiring a liveness

signature in additi on to matched biometric informa-

tion. Methods can incl ude medical measurements such

as pulse oximetry, electrocardiogram, or odor. In a few

cases, liveness information is inherent to the biometric

itself, i.e., the biometric cannot be captured unless the

user is live, e.g., electrocardiogram as a biometric.

While liveness algorithm makes spooﬁng more difﬁ-

cult, they need to be considered as components of a

biometric system which bring with it performance

characteristics, as well as factors such as ease of use,

collectability, user acceptance, universality, spoof-abil-

ity, permanence, and, in some cases, even uniqueness.

No system is perfect in its ability t o prevent spoof-

attacks. However, liveness algorithms can reduce this

vulnerability to minimize the risk of spooﬁng.

▶ Anti-spoofing

▶ Liveness Detection: Fingerprints

▶ Liveness Detection: Iris

▶ Liveness Detection: Fingerprint

Liveness Detection: Fingerprint

STEPHANIE A. C. SCHUCKERS

Clarkson University, Potsdam, New York, USA

Synonyms

Anti-spooﬁng; Vitality

924

Liveness Detection

Definition

In biometric systems, the goal of liveness testing is

to determine if the biometric being captured is an

actual measurement from the authorized, live person

who is present at the time of capture. While ﬁngerprint

systems may have an excellent performance and im-

prove security. Previous studies have shown it is not

difﬁcult to make molds of latent ﬁngerprints left

by legitimate users and to create fake ﬁngers made

from Play-Doh, gelatin, and silicone materials to fool

a variet y of ﬁngerprint scanners, termed spooﬁng.

Liveness detection reduces the risk of spooﬁng by

requiring a liveness signature, in addition to matched

biometric information. Methods can be divided into

hardware and software categories. Hardware methods

include measurements like pulse oximetry, electrocar-

diogram, or odor, while software based measurements

use additional processing of the biometric information

itself to isolate liveness signatures like perspiration and

deformation. While liveness algorithm makes spooﬁng

more difﬁcult, they need to be considered as compo-

nents of a biometric system, which bring with it per-

formance characteristics along with factors such as

ease of use, collectability, universality, spoof-ability,

permanence, and in some cases, even uniqueness. No

system is perfect in its ability to prevent spoof -attacks.

However, liveness algorithms can reduce this vulnera-

bility to minimize the risk of spooﬁng.

Fingerprints are graphical ridge-valley patterns from

human ﬁngers. Fingerprint recognition is a widely

used and efﬁcient technique for biometric authentica-

tion. While ﬁngerprint systems may have excellent per-

formance and improve security, previous studies have

shown it is not difﬁcult to make molds of latent ﬁnger-

prints left by legitimate users and to create fake ﬁngers

made from Play-Doh, gelatin and silicone materials to

fool a variety of ﬁngerprint scanners [1, 2]. The most

famous of which is the work by Matsumoto and collea-

gues. In the reports, two different techniques were used

to create a mold. The ﬁrst technique directly used a

subject’s ﬁnger to create the mold in free molding plastic,

whereas the second technique involved making a mold

from a latent ﬁngerprint image. Casts were made of

gelatin material and termed ‘gummy ﬁngers’. Veriﬁca-

tion rates of gummy ﬁngers ranged from 68 to 100%. For

method of creating a cast from residual ﬁngerprints, all

ﬁngerprint systems were able to enroll the spoof ﬁnger

and verify more than 67% of the attempts. Similar results

have been obtained on subsequent studies with various

materials including silicon, clay, and Play-Doh [1, 2 ],

and one study which looked at cadaver ﬁngers [2].

Currently, International Biometric Group with spon-

sorship from Financial Services Technology Consor-

tium (FSTC) is hosting an effort to conduct spoof

trials with vendor volunteers called SPOOF 2007.

It should be noted that vulnerability to spooﬁng is

not assessed as part of the false accept ratio, a typical

assessment measure of biometric devices. A false accept

is when a submitted sample is incorrectly matched to

a template enrolled by another user. This only refers

to a zero-effort attempt, i.e., an unauthorized user

making an attempt with their own biometric to gain

access to a system. If the false accept ratio is kept low,

then the probability of speciﬁc user with criminal in-

tent matching another template is very low. The false

accept ratio does not give information on the vulnera-

bility of a system to spoof attacks.

Even though biometric devices use physiologic

information for identiﬁcation/veriﬁcation purposes,

these measurements rarely indicate liveness. The goal

of liveness testing is to determine if the biometric being

captured is an actual measurement from the author-

ized, live person who is present at the time of capture.

Overview of liveness approaches are described in

[2–5]. Performance of ﬁngerprint liveness to separate

live and spoof ﬁngers is measured by live false reject

rate and spoof false accept rate. Equal error rate

between these two measures and receiver operating

characteristic curves can also be used as described in

Biometric Security Overview. Marcialis et al., provides

a table which compares datasets used for testing and

performance of liveness approaches.

Methods to measure liveness fall into several cate-

gories. In 2007 [6], a taxonomy is presented whereby

methods are divided into soft ware and hardware-

based. A similar division is suggested, but also consider

an additional category where liveness is inherent to the

biometric, i.e., it must be present in order to capture

the biometric [2]. In the ﬁrst liveness is captured

through additional hardware integrated with the ﬁn-

gerprint sensor. For ﬁrst category software-based tech-

niques, involves further processing of the biometric

signature to obtain liveness information. For example,

this may mean extracting perspiration information

from ﬁngerprint image. The second software based

approach is where liveness is an inherent part of the

biometric, in other words, the biometric cannot be

Liveness Detection: Fingerprint

925

captured unless the subject is alive. An example for this

category is the electrocardiogram, which has been sug-

gested as a biometric [7] and where liveness is inherent

to collection of this biometric. Liveness in most cases is

not inherent to be able to measure a ﬁngerprint bio-

metric. Most systems that consider liveness in ﬁnger-

print do so through additional software or hardware

processing. Electrocardiogram might be considered a

special case as it has been suggested as an additional

measurement to ﬁngerprint recognition so it can be

considered as hardware liveness approach and it may

be potentially inherent to the biometric if the electro-

cardiogram is used as a biometric.

Hardware

The ﬁrst method uses extra hardware to acquire life

signs. Previously developed approaches measure ﬁnger-

tip temperature, pulse, pulse oximetry, blood pressure,

electric resistance, odor, multi-spectral information, or

electrocardiogram (e.g., [8, 7, 9, 10]). These methods

require dedicated hardware integrated with the ﬁnger-

print system. Electrocardiogram is the electrical mea-

surement of the heart collected through elec trodes

on two skin contact points on the body which need

to be on opposite sides of the heart (e.g., two hands,

hand and foot). Pulse oximetry is the measurement

of the oxygen content of the blood through the com-

parison of the absorption of two wavelengths of

light by the blood. This measurement requires a LED

and photodetector on opposite sides of the ﬁnger

and typically needs to be shielded from ambient light.

This absorption also varies as the heart be ats and

can be a measure of pulse, and therefore may require

a few seconds to compute to record one or two com-

plete heart beat cycles. A critical component to hard-

ware-based approaches is how the additional hardware

is integrated with the ﬁngerprint sensor. It should be

integrated in such a way that it cannot be spoofed with

any live ﬁnger in combination with a spoof.

The following paragraph describes two ﬁngerprint

sensors, multispectral and ultrasound, which naturally

capture liveness information. They are placed here in

the hardware category, because these approaches, while

commercially viable, require purchase of a speciﬁc

scanner and are not applicable to standard ﬁngerprint

readers. One commercially available ﬁngerprint sensor

(Lumidigm, USA) uses a multispectral sensor, from

which multiple wavelengths of light and different

polarizations allow new data to be captured, which is

unavailable from a conventional optical ﬁngerprint

reader. Based on the multiple spectral images, they

have developed a spoof detection method [10]. Simi-

larly, ultrasound measurements have been suggested as

a way to measure ﬁngerprint images (Opt el, Poland).

While ﬁngerprint measured by ultrasound might be

able to image a spoof or cadaver ﬁngerprint itself,

using additional information from the ultrasound

measurement would likely be capable of separating

live from spoof images. Both approaches most likely

need additional processing from the ﬁngerprint image

itself to determine liveness.

Software

The second method uses the information already pres-

ent in the ﬁngerprint imag e to detect life signs, for

example, skin deformation, pores, power spectrum,

or perspiration pattern.

Skin deformatio n and elasticity. Skin deformation

technique uses the information regarding how the

ﬁngertip’s skin deforms when pressed against the scan-

ner surface [11–14]. The studies show that when a real

ﬁnger moves on a scanner surface, it produces a signif-

icant amount of non linear distortion. However, fake

ﬁngers are more rigid than skin and the deformation is

lower even if they are made of highly elastic materials.

One approach quantiﬁes this considering multiple

frames of clockwise motion of the ﬁnger [12]. The

performance of this method has an equal error rate

of 11.24% using 45 live subjects and 40 fake ﬁngers.

A study by Zhang et al. [14] uses a thin-plate spline

distortion model over multiple frames, while the ﬁnge r

is moved and resulted 4.5% EER in a dataset of 120

fake ﬁngerprints from silicon from 20 individuals.

Another metho d considers the deformation in a single

image compared to a template [11]. This study

achieved 82% for a small dataset.

Perspiration pattern. Previously, our laboratory has

demonstrated that perspiration can be used as a mea-

sure of liveness detection for ﬁngerprint biometric

systems. Unlike spoof and cadaver ﬁngers, live ﬁngers

demonstrate a distinctive spatial moisture pattern

when in physical contact with the capturing surface

926

Liveness Detection: Fingerprint

of the ﬁngerprint scanner. The pattern in ﬁngerprint

images begins as ‘patchy’ areas of moisture around the

pores spreading across the ridges over time. Image/

signal processing and pattern recognition algorithms

have been developed to quantify this phenomenon

using wavelet and statistical approaches [15–17].

These approaches require two time-series images,

which might not be convenient for the users. Other

methods to quantify this phenomenon have been de-

veloped for a single image [18]. Performance has

achieved approximately 10% live/spoof EER for earlier

papers on a dataset of 80 spoof, 25 cadaver, and 58 live

images to perfect separation in later papers on this

small dataset [16].

Characteristics of spoof and live images. A natural

extension to the speciﬁc categories above is to begin to

assess the characteristics that deﬁne live and spoof

ﬁngers, which cover a broad range [5, 13, 19–21].

Theseincludeimagepowerspectrumthatrevealsstamp

fabrication process [5], noise residue in the valleys

due to spoof material [ 19, 21], and combinations of

multiple factors, for example, fusion of perspiration

and deformation features [13].

Image power spectrum has been considered as an

effective feature for vitality detection [5, 20]. The dif-

ference between live and spoof images is mainly

due to the stamp fabrication process, which causes an

alteration of frequency details between ridge and

valleys. The Fourier transform feature can quantify

the difference in terms of high frequency informa-

tion loss for fake ﬁngers. This approach is tested for

a single scanner and silicone spoof material with

average spoof/live EER of 2.4% on a dataset of

720 fake and 720 live images from 36 individuals [6]

and for gelatin and silicon with an average of 23%

EER for a dataset of 900 fake and 450 live images

from 30 individuals [20].

In other study [13], a sequence of images is used

to measure skin elasticity, but some of the measures

may be capturing perspiration information as de-

scribed above. No special motion is required for

the ﬁnger. They achieve results of 4.78% on a dataset

of 470 spoof images from 47 spoof casts and 300

live images from 15 individuals. In a second study,

fusion of multiple features, two based on perspiration

signal and two based on skin elasticity, was performed

in 2007 [22]. Result showed 4.49% EER on the

same dataset.

Liveness Algorithm Framework

Fingerprint liveness algorithms can fall into types de-

scribed above (hardware, software, and inherent).

Other factors that separate liveness algorithms include

(1) dynamic/static, (2) user training, and (3) binary/

user speciﬁc. Table 1 compares ﬁve ﬁngerprint liveness

algorithms within the context of this framework.



Dy namic or static: Liveness algorithms may require

only one frame or rely on multiple fram es to

measure the dynamic nature of the system to de-

tect liveness [5]. For example, many of the per-

spiration proposed approaches require more than

one image [15], although recent work has used one

image [18]. Other dynam ic approaches are related

to deformation [12–14]. Note that pulse oximetry

do not require multiple ﬁngerprint image frames,

however, they may require more time to record one

or more full hear t cycles.



User training: Some liveness algorithms rely on

speciﬁc user actions to determine liveness. This

may include a procedure (deform ation changes

due to rotating the ﬁnger), which require user

training [12, 14].



Binary (live/spoof) versus user speciﬁc: Liveness

algorithms can be made general across all subjects,

that is, the same algorithm is used for all subjects to

determine liveness producing a binary result: live

or non live (Fig. 1). Other approaches can be made

subject speciﬁc, that is, a liveness algorithm is

Liveness Detection: Fingerprint. Table 1 Liveness

algorithm types and factors

Hardware/

Software

Multiple/

Single

Binary/

User

specific

User

training

Perspiration S M/Si B/US None

Pulse

oximetry

H – B None

Multi-

spectral

H Si B/US None

Deformation S M/Si B UT or

none

ECG H – B/US UT

H Hardware; S Software; M Multiple; Si Single; B Binary; US User

Speciﬁc; UT User Training; – indicates not applicable

Liveness Detection: Fingerprint

927

imbedded as part of the biometric template. For

example, work has been shown for storing a perspi-

ration pattern speciﬁc to an individual [23]. While

not speciﬁcally mentioned for the multi-spectral

ﬁngerprint scanner (Lumidigm, USA), it is possi-

ble that a medical spectroscopy-based liveness

approach could be user speciﬁc. Electrocardiogram

can also be user-speciﬁc, that is, used as a biometric

[7](Fig. 2).

Other characteristics for evaluating biometrics systems,

such as ease of use, collectability, user acceptance, uni-

versality, uniqueness, permanence, and spoof-ability,

need to be considered before implementing a liveness

algorithm. These were described in the Biometric Secu-

rity Overview chapter. Table 2 considers the same live-

ness algorithms from Table 1 within the context of

this framework.



Ease of use: Some liveness approaches may be easier

to use. For example, ﬁngerprint deformation ap-

proach that requires a speciﬁc rotation procedure

may be considered more difﬁcult to use [12, 13].

Lumidigm approach for spectroscopy where live-

ness is collected as part of the biometric collection

itself may be considered easier to use.



Collectability: The hardware, equipment setup, and

relationship to the user impacts the collectabilit y

of the liveness algorithm. For example, approaches

that may be more difﬁcult to collect include the

electrocardiogram, which requires two points of

contact on opposite sides of the body or pulse

oximetry, where the ﬁnger must be enclosed

to protect from ambient light. In comparison,

approaches that use the traditional biometric

Liveness Detection: Fingerprint. Figure 1 Example of

live and non-live fingerprints captured by Capacitive DC

scanner: (a) live ﬁnger; (b) spoof ﬁnger made from Play-

Doh; (c) spoof ﬁnger made from gelatin; (d) cadaver ﬁnger.

Liveness Detection: Fingerprint. Figure 2 Perspiration patterns. Spoof, live, and cadaver patterns are shown from left to

right. The perspiration pattern is the reconstruction of the isolated wavelet coefficients obtained from two fingerprint

images in time, by the algorithm described [4].

928

Liveness Detection: Fingerprint

equipment for measurement of liveness might be

considered easier to collect.



User accept ance: For ﬁngerprint liveness, approaches

with low user acceptance are ones that are more likely

to be linked with medical conditions due to privacy

concerns (electrocardiogram, pulse oximetry, and

multi-spectral) (Fig. 3).



Universality:Obviouslyallauthorizedusers

should be live when presenting their biometric ;

however, the liveness signature may be difﬁcult

to measure in some subjects. For example, perspi-

ration in ﬁngerprint images may b e difﬁcult to

measure in individuals with very dry skin, which

is also a problem with measuring the ﬁngerprint

image itself.



Uniqueness: For liveness approaches, which are

inherent to the biometric, this factor is critical.

However, as mentioned above, electrocardiogram

in combination with ﬁngerprint would not need

uniqueness as a characteristic, whereas, the

electrocardiogram alone may need further research

to address uniqueness [7].



Permanence: Permanence typically refers to perma-

nence of the speciﬁc biometric pattern over time.

Similar to above, this more directly applies to live-

ness approaches, which are inherent to the biomet-

ric, where the biometric/liveness signature may

vary over time. For example, in the initial work

introducing perspiration patterns as a unique

liveness pattern, only 3 months were considered

[23]. It is unknown if these patterns persist beyond

that period. Electrocardiogram may also have difﬁ-

culties with permanence as the electrocardiogram is

impacted by health conditions [7].



Spoof-ability: Spoof-ability considers the possibility

that the liveness mechanism which is put in place to

protect the system from spooﬁng can be spoofed.

For example, in the case of pulse oximetry, it may

be possible to spoof with a clear spoof, which

allows transmission of the light needed to make

Liveness Detection: Fingerprint. Table 2 Liveness algorithm characteristics

Ease of use Collectability

User

acceptance Universality Uniqueness Permanence Spoof-ability

Perspiration H H H M L M M

Pulse

oximetry

LL L H– – H

Multi-

spectral

HH M H – – L

Deformation L L H M – – M

ECG L L L H L H H

H High; M Medium; L Low; –indicates not applicable

Liveness Detection: Fingerprint. Figure 3 Spectral image of the fingerprint, the ring pattern, and the band-selected

frequency analysis from [20].

Liveness Detection: Fingerprint

929