人脸识别算法
TheDesig
a
dImpleme
tatio
ofAlgorithmsforHuma
FaceRecog
itio
1i
f人脸识别算法
摘
要
人脸自动识别是模式识别领域的一项热门研究课题,有着十分广泛的应用前景。本文对人脸位置矫正,人脸的特征提取和识别这些方面进行了研究,并提出了相应的实现算法。人脸位置矫正作为人脸检测定位的一个环节,在计算机人脸识别中具有重要的意义。本文第二章提出了一种基于单人脸灰度图像中眼睛定位的人脸位置矫正方法,它是针对人眼灰度变化特点、人眼几何形状特征及双眼的轴对称性而设计的。实验结果表明,该方法对于双眼可见单人脸灰度图像能实现快速有效矫正,并能在矫正结果中精确给出双眼瞳孔位置。本文第三章提出了一种基于神经网络的主元分析人脸图像识别方法。该方法利用非线性主元分析神经网络对人脸图像提取人脸特征(矢量),并在BP神经网络上实现了对人脸图像的识别。实验结果证明了该方法的有效性和稳定性。
关键词
1ii
f人脸位置矫正,人脸特征提取,人脸识别,神经网络,灰度图像,图像块纵向复杂度,主元分析法,
1iii
fTheDesig
a
dImpleme
tatio
ofAlgorithmsforHuma
FaceRecog
itio
Stude
tYa
gboGu
AdvisorDrWe
mi
gCao
Departme
tofComputerScie
cea
dTech
ologyCollegeofI
formatio
E
gi
eeri
gZhejia
gU
iversityofTech
ology
Abstract
Theautomaticrecog
itio
ofhuma
facesisahotspoti
thefieldofpatter
recog
itio
whichhasawidera
geofpote
tialapplicatio
sAstheresultsofouri
depthresearchtwoalgorithmsareproposedo
eforfaceposeadjustme
ttheotherforfacialfeatureextractio
a
dfaceide
tificatio
Faceposeadjustme
tasaloopofhuma
facelocatio
isveryimporta
ti
computerfacerecog
itio
Chapter2ofthisthesisprese
tsa
ewapproachtoautomaticfaceposeadjustme
to
grayscalestaticimageswithasi
glefaceI
afirststagetherightpositio
sofeyesarepreciselydetectedaccordi
gtoseveraldesig
edparameterswhichwellcharacterizethecomplexcha
gesofthegrayparameteri
a
darou
deyesa
dthegeometricalshapeofeyesDuri
gtheseco
dstagebasedo
thelocatio
a
dthesymmetryfeatureofeyesthei
cli
atio
a
gleiscalculateda
dthefacepositio
isredressedTheexperime
tatio
showsthatthealgorithmperformsverywellbothi
termsofratea
dofefficie
cyWhat’smoreduetothepreciselocatio
ofeyestheapplesoftheeyesaredetected1iv
fI
chapter3a
ovelapproachtohuma
faceimagerecog
itio
basedo
pri
cipalcompo
e
ta
alysisa
d
eural
etworkshasbee
proposedByusi
gBP
eural
etworkshuma
faceimagesaresuccessfullyclassifieda
drecog
izedaccordi
gtotheoutputofBPNNwhosei
putistheeige
vectorextractedfromr