2024年4月23日发(作者：台式电脑怎么安装打印机)

Toappearinthe6th2004.

Multi-biometricsUsingFacialAppearance,ShapeandTemperature

inChen

ComputerScience&EngineeringDepartment

UniversityofNotreDame

NotreDame,IN46556U.S.A.

{kchang,kwb,flynn}@

Abstract

Wepresentresultsofthefirststudytoexamineindividual

andmulti-modalfacerecognitionusing2D,3Dandinfra-

nsorcap-

turesdifferentaspectsofhumanfacialfeatures;appearance

inintensityrepresentingsurfacereflectancefromalight

source,shapedatarepresentingdepthvaluesfromthecam-

era,andthepatternofheatemitted,-

ployadatabasecontainingagallerysetof127imagesand

aPCA-basedapproachtunedseparatelyfor2D,3DandIR,

wefindrank-onerecognitionratesof90.6%for2D,91.9%

for3Dand71.0%ingeachpairofmodal-

ities,wefindamulti-modalrank-onerecognitionrateof

98.7%for2D-3D,96.6%for2D-IRand98.0%for3D-IR.

Whenallthreemodalitiesarecombined,weobtain100%

ultsshowninthisstudyappeartosup-

porttheconclusionthatthepathtohigheraccuracyand

robustnessinbiometricsinvolvesuseofmultiplebiometrics

ratherthanthebestpossiblesensorandalgorithmforasin-

glebiometric.

uction

Faceisoneofthemostimportantandcommonlyusedbio-

metricsforidentificationduetoitsacceptability,universal-

ityandnon-intrusiveness[1].Theidentificationofthehu-

manfacein2Dhasbeeninvestigatedbymanyresearchers,

butrelativelyfewstudiesusingotheraspectsoffacialfea-

tureshavebeenreported.

Eachimagingmodalityhasitsownbenefitsandprob-

lemswhenappliedtofacerecognition.2Dimagesaregen-

ceived

benefitsfromusing3Drelativeto2Ddataincludelessvari-

1

ationobservedduetofactorssuchasmakeupandreduced

sensitivitytoilluminationchanges(eventhougha3Dsens-

ingoperationisinfluencedbytheillumination).Also,the

patternofheatemittedfromthehumanbody(face)may

effectivelybeconsideredasacharacteristicofeachindivid-

ual[2].Thispaperfirstaddressestheeffectivenessofeach

individualbiometricsource,usingaPCA-basedtechnique

toinvestigatetheidentificationaccuracyoftheapproach.

Then,weconsiderthecombinationofdifferentfacialfea-

thefirststudytoexaminethesethreeface

biometricsandtocomparethedifferentpairsoffacebio-

ougheachimagingmodalityhasitsown

advantagesanddisadvantagesdependingoncertaincircum-

stances,thereisoftensomeexpectationthat3Ddatashould

r,norigorousexperimen-

experimentsreportedinthisstudyareaimedat(1)testing

thehypothesisthatthereexistsasuperiorityofaccuracyfor

onebiometricoverotherbiometricsources,usingthePCA-

basedmethod,and(2)exploringwhetheracombinationof

2D,3DandIRfacedatamayprovidebetterperformance

ectofcombiningdif-

ferentbiometricsishowtocombineresultsprovidedby

individualsourceseffectivelyduringthedecisionprocess.

Manydifferentapproachescouldbeenvisionedforcom-

biningmultipletypesofbiometricinformation[3,4,5].In

general,theycanbethoughtofasoccurringattheimage

level,themetriclevel,study,we

term“multi-modalbiometrics”isusedheretorefertothe

useofdifferentsensortypeswithoutnecessarilyindicating

antaspects

ofsomerelatedmulti-modalstudiesaresummarizedinTa-

ble1.

Table1:Previousstudiesofdualbiometricsusingface

Source

(Year)

Chang

(’03)

[6]

Biometric

Source

(Subjects)

2Dface&

3Dfacial

shape(275)

sum

Fusion

Level

hybrid

(rank&

metric)

Wang

(’02)

[8]

2Dface&

3Dfacial

shape(50)

weighted

sum

metric

Liu

(’99)

[10]

2Dface&

3Dfacial

shape(200)

sum

feature

sum,

decision

tree

Frischholz

(’00)

[13]

2Dface,

voice&lip

motion(180)

sum,min,

product,

max,median

hybrid

(rank&

metric)