基于视频的人脸识别研究进展
886
计算
机学报
2009年
[51]YanSC,XuD,ZhangB,ZhangHJ.Graphembeddingand
extensions
z
ComputerVisionandImageUnderstanding,2003,91(1);
188’。213
Ageneralframeworkfordimensionalityreduc。。Transactions
on
tion.IEEEPattern
AnalysisandMachine[58]ChoudhuryA。ClarksonB,JebaraT,PenlandA.Multimo—dal
personrecognitionusing
Intelligence,2007.29(1):40 51
unconstrained
on
audioandvideo//Video-based
[523
Gross
R,ShiJ.TheCMUMotionofBody(MoBo)data—ProceedingsoftheConference
Audio‘and
base.RoboticsInstitute,CarnegieMellonUniversityl
Tech—nicalReportCMU—RI—TR-01-18,2001
[53]TeferiD,BigunJ.Damascening
video
databasesforevalua一
tion
offacetrackingandrecognition--TheDXM2VTSdata—base.PatternRecognitionLetters,2007,28(15)l2143—2156
[54]ZhangY。Martinez
A
M.Aweightedprobabilisticapproach
to
face
recognition
frommultipleimagesandvideosequences.
ImageandVisionComputing,2006,24(6):626—638
[55]Al—AzzehM。EleyanA.DemirelH.PCA-basedface
recogni—
tion
from
video
using
super-resolution//Proceedingsof
the
23rdInternationalSymposiumon
Computer
andInformation
Sciences.Istanbul,2008:1-4
[56]GokselD.Exploitingspace—time
statistics
ofvideosforface
h11ucination[Ph.D.dissertation].CarnegieMellon
Univer一
sity。Pittsburgh,USA,2007
[57]ChowdhuryA,ChellappaR.Face
reconstruction
frommo-tionalnocularvideo
using
uncertainty
analysisand
a
generic
model.
YANYah,bomin1984,Ph.D..
Hismainresearchinterestsfocus
on
pat
tern
recognitiomBackground
ThisworkissupportedbytheNationalNaturalScience
FoundationofChinaundergrantNo.60872084andtheSpe—cializedResearchFundfortheDoctoralProgramof
Higher
Educationunder
grant
No.20060003102.Traditionalstillimage-basedfacerecognitionhasachieved
great
SUCCP嬉S
in
constrainedenvironments.However,once
the
conditions,including
illumination,pose,expression,
age,etc.,change
tOO
much,theperformancedeclinesdra—matically.The
recent
FRVT2002showsthattherecognition
performanceoffaceimagescapturedin
an
outdoorenviron—mentanddifferentdays
is
still
not
satisfying.Currentstill
image—based
face
recognitionalgorithms
are
even
faraway
fromthecapabilityofhumanperceptionsystem.Ontheoth—
er
hand,psychologyandphysiologystudieshaveshownthat
motion
can
helppeopleforbetterfacerecognition.万
方数据BiometricPerson
Authentication.WashingtonD.C,1999
l
176—180
[59]ZhangZ
Y,Liu
zC,Adler
D,Cohen
MF,HansonE,Shan
Y.Robustandrapidgenerationofanimatedfacesfromvideo
imageslAmodel—based
modeling
approach.International
JournalofComputerVision,2004,58(2)I93—119
[60]ZhouX,BhanuB.Integratingfaceand
gait
forhuman
recog—
nition
ata
distanceinvideo.IEEETransactions
on
Systems,
Manand
Cybernetics,PartB,2007,37(5):1119—1137
[61]JingXY,YaoY
F,ZhangD,YangJY,LiM.Faceand
palmprintpixellevelfusionandkernel
DCV-RBF
classifier
forsmallsamplebiometricrecognition.PatternRecognition,
2007.40(11):3209—3324
[62]
YanY,ZharlgYJ.Multimodalbiometricsfusionusingcot-
relationfilterbank//Proceedingsofthe19thIAPRInterna一
Conference
on
PatternRecognition.Tampa,2008,
MoBT7.3(卜4)
ZHANGYu-Jin,bornin
1954,Ph.D。,professor。
Ph.D.supervisor.Hismainresearchinterestsincludeimage
engineering(imageprocessing,image
analysis,imageunder-
standingandtechnique
application).http://www.ee.tsing-
hua.edu.cn/~zhangyujin/
During
the
past
several
years,many
researchefforts
have
been
concentratedon
video-based
face
recognition.
Comparedwithstillimage—basedfacerecognition,truevideo-
basedfacerecognitionalgorithmsthat
use
bothspatialand
temporalinformationstartedonlya
few
years
ago.Nocom—
prehensive
survey
inthisfieldhasbeen
made,and
a
lotofis—sues
invideo-basedface
recognitionstillhave
not
beenad—dressedwell.Sothe
content
ofthispapergives
an
overview
ofthemostexistingmethodsinthefieldofvideo-basedface
recognition.Asuitable
classification
fordifferentmethods
hasbeenmade,therespectivepros
and
cons
oftypicaltech’
niquesineachmethodgroup
are
analyzed.Theimportantis—
sues
whichneedto
besolved,theprospectsforfuturedevel—
opmentandsomesuggestionsforfurtherresearchworksare
putforward
to
meetthegoalofthis
paper.