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

Подождите немного. Документ загружается.

Protocol

Protocol is the preset procedure by which each tem-

plate in the database is compared to some speciﬁc

feature during the performance evaluation of the ﬁn-

gerprint-matching approach, and the rule to treat the

failed matching.

▶ Fingerprint Matching, Automatic

Pseudo-Random Number Generator

An algorithm to generate a sequence of numbers that

approximate the properties of random numbers. The

sequence is not truly random since it is completely

determined by a few parameters and initial values.

Pseudo-random numbers are useful for cryptography

and simulation.

▶ SFinGe

Psychology of Gait and Action

Recognition

FRA NK E. POLLICK

Department of Psychology, University of Glasgow,

Glasgow, UK

Synonyms

Action categorization; Action understanding; Biologi-

cal motion perception

Definition

The psychology of gait and action recognition strives

to understand the processes that underlie how people

detect, recognize and understand the movements of

others. Since gait is a fundamental human activity,

it has formed an impor tant visual signal for psycholo-

gists to examine. Experiments have shown that sparse

representations of gait support the recognition of iden-

tity, gender, and emotion by obser vers even when

viewing conditions are degrade d. The study of gait

and action recognition focuses on several questions,

including: what visual properties uniquely specify

human movement; how to quantify human perfor-

mance in action recognition; and the neural mechan -

isms that form the basis of decoding human

movement.

Introduction

The modern study of the psychology of human move-

ment, in particular the perception of gait, star ts with

the work of the Swedish Psychologist Gunnar Johans-

son in the 1970s [1]. The work of Johansson and his

contemporaries focused on how humans use motion

to infer the structure of objects moving in the world.

To demonstrate this capability, he attached lig hts to the

joints (elbow, shoulder, ankle, etc.) of an actor and

ﬁlmed the actor moving about in a darkened room.

In any indiv idual frame of the movie, the points did

not convey a strong impression of structure. However,

when the movie was played, a vivid impression of the

actor moving through space was obtained (Fig. 1).

These displays of human activity are called

▶ point-

light displays and the general ﬁeld of studying how the

individual point motions spontaneously organize into

the percept of a moving body is known as

▶ biological

motion perception [2].

There are several reasons why point-light displays

form a key contribution to the psychology of under-

standing human actions. The ﬁrst is that point-light

displays represent an action as just the 2D locations of

a set of joint locations on the body and thereby remove

a multitude of other visual information that can be

conveyed by things like hairstyle, clothes, facial expres-

sion, and other factors; thus the contribution of mo-

tion itself can be effectively isolated. A second reason

is that the relevance of motion is highlighted, since

for these displays any particular static frame typically

does not elicit a strong impression of a body. A third

reason is that the motion properties of a small set of

points can be easily quantiﬁed, allowing for biological

motion displays to be compared in exp eriments

against other motion patterns with identical motion

1100

Protocol

statistics. Finally, point-light displays of biological mo-

tion provide a compelling demonstration of the use of

motion to perceptually organize a display, the precise

mechanisms of which are still unclear.

Oneissuewiththeuseofpoint-lightdisplaysis

how to evalu ate the role of form perception. It could

be assumed that the mechanisms beh ind the percep-

tual organization of the p oint-li ght displays pro-

vide

▶ structure-from-motion information regarding

form [3]. However, this form information is not nec-

essarily equivalent to what might be available from

that presented directly from an image or even some

other reduced form such as a stick ﬁgure o r a silhou-

ette. The distinction between form and motion is

important from the psychological perspe ctive since

data from neuroscience suppor t the id ea that the

human visual system is segregated into largely distinct

pathways that specialize i n processing form and mo-

tion information.

The study of point-light displays has been critical

in developing the understanding of how motion can

be used to convey the presence of an actor from

minimal information, and to a degree psychological

research has focused in this domain. However, o ne

other question which has drawn attention is just what

person properties can be derived from the p oint-light

displays. Namely, can actor qualities such a s identity,

gender, emotion, attractiveness, and intent be i den-

tiﬁed from suc h displays? Experimental results gen-

erally indicate that human observers can identify

such qualities at better than chance, though what

cues they use and how to evaluate t heir performance

on an absolute scale are presently areas of active

researc h.

Psychological Studies into Perceiving

Biological Motion and Recognizing

Person Properties

In this section empirical investigations into biological

motion perception and action recognition is reviewed.

The majority of research into biological motion percep-

tionhasinvolvedpoint-lightdisplaysofgait.Researchinto

action recognition has typically also used point-light dis-

plays, though sometimes limited the visual display to

those points that change substantially for the different

actions.Anothermethodological differenc ebetween stud-

ies of biologicalmotion perception and actionrecognition

has been that research into biological motion perception

has typically relied upon psychophysical analyses of the

ability of observers, under normal and degraded viewing

conditions, to detect the presence of a walker or to dis-

criminate the walking direction [2]. In contrast, research

into action recognition has used a variety of experi-

mental techniques aimed at uncovering the underlying

features used by observers to recognize the action being

performed as well as properties of the actor.

Numerous e xp eriments have shown that the abi l-

ity of the perceptual system to detect the biological

motion of a walker is surprisingly resistant to distor-

tions of the walker or the embedding of the walker

in v isual noise. For instance, limiting the lifetime of

the points on the walker or displacing them to points

on the skeleton, as opp osed to joint locations, barely

diminishes the ability of an obser ver to detect a point

light walker. Furthermore, masking the motion of

the points using a background of randomly moving

noise dots still does not greatly reduce the impression

of a human walker unless the masking noise is used in

Psychology of Gait and Action Recognition. Figure 1 Examples of frames taken from an image sequence of a

point-light display of a ballet dancer. Each individual frame is seen predominately as just a collection of points, although

a static human form is possibly visible in some frames. However, a vivid impression of the action is appreciated nearly

instantaneously when all the frames of the sequence are presented consecutively in a motion sequence.

Psychology of Gait and Action Recognition

1101

combination with disruptions to the synchronization

between points on the walker [4]. This disruption

to synchronization, by introducing time delays and

advances among the points of the walker, renders

local motion cues ineffective for detecting biological

motion or discriminating the direction of motion,

forcing the perceiver to rely on global or conﬁgural

cues [5]. One effective way to mask the motion of a

walker is to take the walker points themselves as the

source of points to use as background noise. However,

even when this is done and the masking dots contain

local motion signals identical to those of the walker,

large numbers are required to diminish the impression

of a human walker [6].

The apparent ﬁne tuning of the perceptual system

to point-light displays of walking has raised the ques-

tion of whether or not specialized motion detectors are

involved in the processing of biological motion [7].

Research comparing the perception of biological mo-

tion to other kinds of motion has revealed differences

in motion tuning character istics. Namely, that the pro-

cessing of biological motion involves the integration

of motion information over a larger spati al extent and

a longer temporal window than that found for other

types of motion [8]. However, these results fall short of

proving that dedicated biological motion detectors

exist since they reﬂect the output of the entire action

processing system, which might include specialized

higher-order mechanisms for processing human act-

ions that augment standard motion detectors. Evi-

dence for the involvement of hig her level factors

comes from the breakdown of biological motion de-

tection when the local form and motion relations are

preserved, but the entire display is inverted [9]. Per-

ception of these inverted point-light walkers is im-

paired relative to upright walkers, independent of the

location of the source of gravity [10].

Investigations of action recognition from point-

light displays have shown that a variety of actor prop-

erties and action styles can be reported above chance

[2, 11]. Importantly, for the ﬁeld of biometrics, it has

been shown that human observers can recognize iden-

tity from po int-light displays of gait [12, 13, 14, 15].

The work of Stevenage [12] also compared recognition

of identity from point-light displays to video record-

ings of the same actor s under full light and diminished

light conditions. It was found that identiﬁcation per-

formance was equivalent between the different viewing

conditions and this was taken as strong evidence that

the motion cues contained in gait were sufﬁcient to

provide cues to identity. Further evidence for the im-

portance of motion cues comes from results which

show that even when size and walking frequency are

made equal for all the targets to be recognized, perfor-

mance decreases but recognition of identity is still

greater than chance a nd generalizes to novel viewing

directions [16].

Other actor properties which can be recognized from

point-light displays of gait include gender , emotion and

even vulnerability [17]. Emotion can also be obtained

from point-light displays of whole-body dance move-

mentsaswellasjustthearmperformingeveryday

movements such as knocking. In sports a variety of

athlete characteristics and movement intentions can be

gleaned from observing the action [18]. As might be

expected from the variety of scenarios discussed,

there is not a speciﬁc single action feature that has

been found to explain the recognition of actor properties

and action styles. However, researchers have generally

distinguished between form and motion cues. For exam-

ple, in gender recognition, experiments have focused

on the diagnosticity of form cues encoded in the differ-

ent relative sizes of hips and shoulders, while other

studies have concentrated on differences of hip motion

[19]. In general, both form and motion featu res appear

effective to inform recognition and given the complex-

ity of human motion it is hard to tease apart the

different sources even when using point-lig ht displays.

Even with simple actions and extremely reduced

point-light displays there is a complex pattern of

body postures that unfold in time, and it is an open

question as to what features within this signal are

crucial. One way this complexity has been addressed

in cases such as gender recognition has been to reason

from ﬁrst principles about what features drive recogni-

tion, and to use carefully manipulated action displays

to test hypotheses about these features. However, an-

other approach has been to use techniques of automat-

ic pattern classiﬁcation to quantify how information in

the point-light displays are used for recognition. An

issue with automatic pattern classiﬁers is that while

they can effectively categorize action styles they do

not necessarily provide intuition into what speciﬁc

features differentiate the styles. For this reason they

have been applied in two domains that do not require

an intuitive understanding of the features. One of

these is to quantify levels of human performance and

the other is to invert the pattern classiﬁers so that

1102

Psychology of Gait and Action Recognition

recognizable differences in action style can be injected

into movements.

The use of pattern classiﬁers to quantify human

performance in recognition has been achieved by

using classiﬁers as the standard of comparison for

human performance. If the classiﬁer can be shown to

optimally achieve recognition by using all the available

information then the efﬁciency of human performance

can be expressed as the percentage of available infor-

mation used by the human observers. If, however, it

cannot be shown that the classiﬁer optimally uses all

the information available then it is still possible to use

the classiﬁer to estimate an upper bound of human

efﬁciency or to compare human recognition of differ-

ent action properties against a standard classiﬁer [20].

For the case of recognizing gender from point-light

displays of gait it has been shown that the average

percentage correct in gender identiﬁcation is 66%,

which is moderately above chance of 50%. Efﬁciency

at gender recognition, calculated relative to a model

emphasizing structural features [21] is 26%, which is a

high value since efﬁciency values of 10% or higher are

generally considered excellent performance. This low

percentage and high efﬁciency reﬂects either that the

male and female distributions are highly overlapping

and that humans are very effective in using the avail-

able structural information, or that since the structural

features do not incorporate motion information that

the efﬁciency results are inﬂated [22]. In summary, the

calculation of efﬁciency provides a valuable tool to

examine the recognition of human movement and

provides a means to use me thods of automatic pattern

classiﬁcation or to examine how performance relates to

the modeled use of a speciﬁc feature.

Another application of automatic pattern classiﬁers

has been to ‘‘invert’’ their performance so that instead

of recognizing actions they are injecting style into

normal movements or otherwise modify ing the move-

ments [23, 24]. The intuition behind this is that the

action of a point-light display can be speciﬁ ed by the

three-dimensional coordinates of the body sampled

many times per second, resulting in thousands of

values representing even a simple action. Each action

can be considered as a point in this high dimensional

space and the different styles of the action as different

regions of this movement space. By obtaining classi-

ﬁers to identify these different regions, possibly with

the use of dimensionality reduction techniques, one

is effectively isolating the differences between a stylistic

and a neutral movement. Thus, by inverting the

computational machinery used to recognize the move-

ment one can obtain the ability to synthesize new

movements which contain the features compatible

with the desired style.

Computational and Biologically Inspired

Models of Action Perception and

Recognition

Early models of biological motion processing were

closely tied to the point-light displays of Johansson.

These models took as their input the image coordinates

of the body points in successive frames and attempted

to solve a series of equations for the three-dimensional

structure of the point lights. The operation of these

algorithms was essentially to incorporate the image

coordinates within constraints such as the planarity of

groups of points, or the hierarchical structure of points.

These structure-from-motion calculations were essen-

tially data driven (i.e., not requiring any information

about body structure except for that incorporated into

the computational constraints) and provided a means

to explain both perceptual organization as well as the

perception of body structure. While later empirical

work called into question these particular models

[25], they are still appealing in their approach to

simultaneously explain perceptual organization and

recovery of body form.

The biologically inspired models have taken as a

starting point that the human visual system appears

to separate the processing of motion and static form

early in the processing streams. Additionally, these

largely independent streams appear to converge in a

brain region, known as the posterior superior temporal

sulcus (STS), that brain imaging studies [26] have

shown to specialize in the processing of biological

motion [27]. The modeling approaches have studied

the instantiation of biologically plausible computa-

tions within a hierarchical processing framework of

form and motion [28] or emphasized the potential

for template matching mechanisms to organize point-

light displays [29]. While these biologically inspired

computational models are broadly consistent with

human behavioral experiments they are exceedingly

complex to test at the physiological level. However,

current investigations of the responses of single cells

are beginning to reveal how motion and form neurons

Psychology of Gait and Action Recognition

1103

are organized in cortex and the form and motion

image characteristics to which they respond [30].

The computational and biologically inspired

models have focused on the early and mid levels of

visual processing in the interpretation of biological

motion. However, since obtaining a v isual understand-

ing of the actions of others has signiﬁcant social signif-

icance there has been activity in trying to understand

how deeper meanings such as goals and intentions of

actions are recovered. While it is possible that this

understanding arises simply from a visual matc hing

process that involves increasingly elaborate representa-

tions of the visual signal, there is evidence that a direct-

matching route works by directly mapping visual

input into ones own behavioral repertoire of actions.

These direct-matching models are largely inspired

by the ﬁnding of brain networks that represent

both the production and perception of goal directed

actions [31]. Consistent with these models recent brain

imaging experiments have found the functional

representations of movement goals and movement

kinematics to be differentially represented within

these networks [32].

Computational models of biological motion have

proven useful in many ways. Not only do they provide

a compact means to express how recognition might

occur but they often lead to testable hypotheses that

can be explored with further experiments. They also,

importantly, allow a common framework for describ-

ing biological motion perception that can span related

efforts in neuroscience and experimental psychology to

understand how actions are recognized.

Related Entries

▶ Evaluation of Gait Recognition

▶ Gait Recognition, Model-Based

▶ Human Detection and Tracking

▶ Gait Recognition, Motion Analysis for

▶ Surveillance

References

1. Johansson, G.: Visual perception of biological motion and

a model for its analysis. Percept. Psychophys. 14(2)(Oct),

201–211 (1973)

2. Blake, R., Shiffrar, M.: Perception of human motion. Annu. Rev.

Psychol. 58, 47–73 (2007)

3. Ullman, S.: The Interpretation of Visual Motion. MIT Press,

Cambridge, MA (1979)

4. Hiris, E., Humphrey, D., Stout, A.: Temporal properties in mask-

ing biological motion. Percept.Psychophys. 67(3), 435–443 (2005)

5. Lu, H., Liu, Z.: Computing dynamic classiﬁcation images from

correlation maps. J. Vision 6(4), 475–483 (2006)

6. Thornton, I.M., Pinto, J., Shiffrar, M.: The visual perception of

human locomotion. Cognit. Neuropsychol. 15(6–8), 535–552

(1998)

7. Mather, G., Radford, K., West, S.: Low-level visual processing of

biological motion. Proc. R. Soc. London, Ser. B – Biol. Sci. 249

(1325), 149–155 (1992)

8. Neri, P., Morrone, M.C., Burr, D.C.: Seeing biological motion.

Nature 395(6705), 894–896 (1998)

9. Pavlova, M., Sokolov, A.: Prior knowledge about display inver-

sion in biological motion perception. Perception 32(8), 937–946

(2003)

10. Shipley, T.F.: The effect of object and event orientation on

perception of biological motion. Psychol. Sci. 14(4), 377–380

(2003)

11. Shipley, T., Zacks, J. (eds.): Understanding Events: How Humans

See, Represent, and Act on Events. Oxford University Press,

Oxford (2008)

12. Stevenage, S., Nixon, M., Vince, K.: Visual analysis of gait as a

cue to identity. Appl. Cognit. Psychol. 13, 469–474 (1999)

13. Loula, F., Prasad, S., Harber, K., Shiffrar, M.: Recognizing people

from their movement. J. Exp. Psychol.: Human Percept. Per-

form. 31, 210–220 (2005)

14. Troje, N.F., Westhoff, C., Lavrov, M.: Person identiﬁcation from

biological motion: Effects of structural and kinematic cues.

Percept. Psychophys. 67 , 667–675 (2005)

15. Cutting, J., Kozlowski, L.: Recognizing friends by their walk: Gait

perception without familiarity cues. Bull. Psychonom. Soc. 9,

353–356 (1977)

16. Troje, N.F., Westhoff, C., Lavrov, M.: Person identiﬁcation from

biological motion: Effects of structural and kinematic cues.

Percept. Psychophys. 67 (4), 667–675 (2005)

17. Gunns, R.E., Johnson, L., Hudson, S.M.: Victim selection and

Kinematics: A point-light investigation of vulnerability to attack.

Journal of Nonverbal Behavior 26(3), 129–158

18. Abernethy, B., Gill, D.P., Parks, S.L., Packer, S.T.: Expertise and

the perception of kinematic and situational probability informa-

tion. Perception 30(2), 233–252 (2001)

19. Johnson, K.L., Tassinary, L.G.: Perceiving sex directly and indi-

rectly – meaning in motion and morphology. Psychol. Sci. 16

(11), 890–897 (2005)

20. Pollick, F., Lestou, V., Ryu, J., Cho, S.B.: Estimating the efﬁciency

of recognizing gender and affect from biological motion. Vision

Res. 42(20), 2345–2355 (2002)

21. Cutting, J.E., Profﬁtt, D.R., Kozlowski, L.T.: A biomechanical

invariant for gait perception. J. Exp. Psychol.: Human Percept.

Perform. 4(3), 357–372 (1978)

22. Pollick, F., Kay, J., Heim, K., Stringer, R.: Gender recognition

from point-light walkers. J. Exp. Psychol.: Human Percept. Per-

form. 31(6), 1247–1265 (2005)

23. Troje, N.F.: Decomposing biological motion: A framework for

analysis and synthesis of human gait patterns. J. Vision 2(5),

371–387 (2002)

1104

Psychology of Gait and Action Recognition

24. Brand, M., Hertzmann, A.: Style machines. In: SIGGRAPH 2000

Conference Proceedings, pp. 183–192. ACM, New York (2000)

25. Bertenthal, B.I., Pinto, J.: Global processing of biological

motions. Psychol. Sci. 5(4), 221–225 (1994)

26. Grossman, E., Donnelly, M., Price, R., Pickens, D., Morgan, V.,

Neighbor, G., Blake, R.: Brain areas involved in perception of

biological motion. J. Cognit. Neurosci. 12(5), 711–720 (2000)

27. Oram, M.W., Perrett, D.I.: Responses of anterior superior tem-

poral polysensory (stpa) neurons to biological motion stimuli.

J. Cognit. Neurosci. 6 (2), 99–116 (1994)

28. Giese, M., Poggio, T.: Neural mechanisms for the recognition

of biological movements.. Nat. Rev. Neurosci. 4(3), 179–192

(2003)

29. Lange, J., Lappe, M.: A model of biological motion perception

from conﬁgural form cues. J. Neurosci. 26(11), 2894–2906 (2006)

30. Vangeneugden, J., Pollick, F.E., Vogels, R.: Functional differenti-

ation of macaque visual temporal cortical neurons using a para-

metric action space. Cerebral Cortex Advance Access published

on July 16, 2008

31. Rizzolatti, G., Craighero, L.: The mirror-neuron system. Annu.

Rev. Neurosci. 27, 169–192 (2004)

32. Lestou, V., Pollick, F.E., Kourtzi, Z.: Neural substrates for action

understanding at different description levels in the human brain.

J. Cognit. Neurosci. 20(2), 324–341 (2008)

Punch-in, Clock-in, Punch-out,

Clock-out, Punch

The term punch, describes the act of a mechanical

strike putting a hole, or ‘‘punching’’ the timecard to

signify the employees action of either registering for

work, or leaving work. In an electronic world, this has

changed to ‘‘clock-in’’ and ‘‘clock-out’’.

▶ Time and Attendance

Punctum Lacrimale

The punctum lacrimale is located at the corner where

upper eyelid comes together with lower eyelid. It is

prominent in many subjects and appears as a ‘‘D’’

shape. The function of punctum lacrimale is to secrete

tears to keep proper moisture on the surface of eyeball.

▶ Iris Super-Resolution

▶ Automatic Classiﬁcation of Left/Rig ht Iris Images

Pupil

The pupil is a hole in the center of the iris that controls

the amount of light entering the eye.

▶ Iris Image Data Interchange Formats, Standardization

Pupil Phase Engineered Iris

Biometrics

▶ Wavefront Coded

Iris Biometric Systems

Pupil Phase Engineering

A general framework for the design of pupil phase

masks for certain computational imaging systems,

where high-quality image acquisition is addressed

from an optimization pers pective. Extending the

depth of ﬁeld is but one requirement of image quality,

others being controlling and minimizing the impact of

aberrations, motion blur, and scattering from the im-

aging medium, to name a few.

▶ Wavefront Coded

Iris Biometric Systems

Purkinje Images

When illuminating the eye the exterior and interior

surfaces of the cornea and lens reﬂect the illuminating

light forming bright reﬂections within recorded images.

These are known as the Purkinje images. W ithin the ﬁeld

of biometrics Purkinje images are evident within iris

images and can obscure areas of iris texture. Iris capture

systems aim to minimize their effect reducing them in

size and restricting them to within the pupil region.

▶ Simultaneous Capture of Iris and Retina f or

Recognition

Purkinje Images

1105

Quadrant

In Cartesian coordinate system, the intersection of the

two axis (x and y) creates four regions, called quad-

rants. Conventionally, quadrants are labeled counter-

clockwise starting from the upper right (‘‘northeast’’)

quadrant. In the ﬁrst quadrant, both x and y coordi-

nates are positive. In the second quadrant, x-coordi-

nates are negative and y-coordinates are positive. In

the third quadrant, both coordinates are negative and

in the fourth quadrant, x-coordinates are positive and

y-coordinates are negative.

▶ Iris Recognition, Overview

Quality-dependent Fusion

▶ Fusion, Qualit y-Based

Quantum Efficiency (QE)

For imaging sensors, the probability that a single pho-

ton impinging on a detector will be detected by the

sensor. The exact deﬁnition can vary from vendor to

vendor, e.g., some vendors report the QE for photons

that hit a sensor pixel, ignoring the ﬁll factor – the

fraction of the sensor that is actually covered by pixels.

▶ Iris Device

2009 Springer Science+Business Media, LLC

Radiometric Calibration

A process for achieving a direct relation between the

value at a pixel and the absolut e amount of thermal

emission from the corresponding physical scene

element.

▶ Face Recognition, Thermal

▶ Image Formation

RAIC

▶ Iris Recognition at Airports and Border-Crossings

Random Forgery

In signature veriﬁcation, random forgeries (also

known as simple forgeries) represent the case where

forgers claim to be another user but use their own

signature.

▶ Signature Databases and Evaluation

Range Scans

Data that has the 3D depth information of every scanned

point. A scan is the reading of the information for a

typically prespeciﬁed region. A range image is a collec-

tion of pixel values with corresponding depth informa-

tion. In most instances, the sensor used to obtain the

range image is calibrated allowing us to give the distance

measures in physical units such as meters.

▶ Face Recognition, Component-Based

RASTA-Filtering

RASTA-ﬁltering was originally introduced in connec-

tion with perceptual linear prediction (PLP) [4] type of

preprocessing; i.e., band-pass ﬁltering in the log spec-

tral domain. It aims to suppress slow channel varia-

tions assumed to be additive. This ﬁltering principle

has also been applied to cepstral feature based prepro-

cessing [21] in both the log spectral and the cepstral

domains. A general RASTA ﬁlter is deﬁned by:

TzðÞ¼

n¼0

n 

n2



1  px

1

; ð1Þ

where, the numerator is a regression ﬁlter of odd order

N and the denominator is a leaky integrator. A simple

variant of RASTA-ﬁltering is a sliddly window mean

subtraction technique, which c orresponds to a moving

average ﬁlter. Filtering is normally performed in the ceps-

tral domain (CMS). The mean corresponds to the long-

term cepstrum and is normally computed on the speech

part. A silence/speech detector is thus necessary .

▶ Session Effects on Speaker Modeling

Raw Finger Vein Image

Raw ﬁnger vein image is the original infrared ﬁnger

vein image captured by a ﬁnger vein reader. It is

2009 Springer Science+Business Media, LLC