ѧ
JOURNAL OF SOFTWARE
2000Vol.11No.3P.372-378



ߵԶʶо
¸塡꿵Ҷ
ժҪһֻߵĿʶ𷽷.÷ȶԲ̬ͬʹСжλ͹һ,ȻڹһͼϽ.Ϊ,øҶӶ߽нȡؼ,õı.ʹô˷1300ͼϽʵ,˷ٶȺ׼ȷȷ涼нϺõ,Ҷв̬ͬͱʶһ³.
ؼʡԶʶһͶӰߣҶ.
ͼšTP391
Automatic Human Face Recognition Based on Eigencurves
ZHANG Gao-qingWANG Shen-kangYE Mao-dong
(Department of Computer Science and Engineering Zhejiang UniversityHangzhou310027)
AbstractA fast method for face recognition based on eigencurves is presented in this paper.First all faces with different poses and sizes are located and normalized.Then several kinds of eigencurves are built to describe characteristics of faces.And finally,the dominant characteristics are extracted which are composed of representative vectors of faces through analyzing the eigencurves using Fourier describer.This method has been tested on 1300 facial images and shown a good performance of speed and accuracy,and the result also shows it is robust for various head poses and expressions.
Key wordsAutomatic face recognition,face normalization,integral projection,eigencurve,Fourier describer.
һ޶,ʹһ˫̥,沿ҲһĳĲ.Ȼڱ顢͵ȷ˾޴仯¶Ժѵؼʶ,ҪһܹȫԶʶϵͳȴǷǳѵ,漰¼⣺(1) һģʽΪ,(2) 沿,(3)ķ.ܹʵܵϵͳй㷺Ӧǰ,ɷƬʶ𡢶ʧͯϢȫϵͳÿȷϵ.ڹȥ20,Զʶڶо,ĿǰȫЧϵͳδƳ.
еԶʶ𷽷ϿɷΪ1,2ݣġھֲķġȫķ.ǰ߳ȡɢľֲΪ,ȻΪñ׼ģʽʶʶ.Ϊһȫֵ,ͨģƥ,صҪǣԪ(PCA)3ݡ(BP)㷨4Լֵֽ(SVD)5ݵ.Щһ㶼Ҫϴļһضصʵ,Ҷ̬Сǳ.磬͵PCAʹ漸ȱ㣺(a) 㷨KL任,ٶһ΢ܣ(b) ûѵ,ʶʽϵͣ(c) ̬仯.
,Ʋʵһµʶ𷽷(eigencurve,ECV).ECV߽ͨʶ,Ҫصǣ(a) ̬С޹,仯ǳʱӰʶ;(b) ٶȿ(ÿͼԼ5)(c) ѧϰ,ϵͳӶ,ͬʱǿϵͳ³.ϵͳֻٶͼֻһ,ұΪ.ECVļͼ1ʾ.

Fig.1face recognition process
ͼ1ʶͼ
1 һ
ԴСͬڲͬб̶ȵͼйһǷǳҪ,ҹһĽֱӰ쵽ʶЧ.ڱϵͳ,һͬʱҲԶλ,˼ǻһԵʵеͼ,ͷ۾Էֿ,ˣֻҪͼܱܺõطָ,Ϳɽ۾׼ȷؼ,ͬʱĶλ,,׼ɽ׵صõ.һͼСΪ128192,ͬһˮƽλ.
1.1ֵָ
,ͼI(x,y)ֵָ,ʹͷ۾Էֿ.ϵͳٶͼֻһ,ָԼ.ͼ2(b)ͼĵֱͼ,ԵķֱֵӦİ(ͷ۾üë)(ͷյ).ҵ1ֵӦӦİ,Ϊһ㱳ǻİҪ,,ֻҵǰֵ,ȡһҶֵΪֵ.ڴڶѡֵ,ͨԶֵʹͼĶ̷ֵǳɿ.


Fig.2 The thresholding of input images
ͼ2ԭʼͼֵָ
ֱͼгͼĻҶȷֲŵصӰһЩƷŵ,ǵĴڻᵼ¶Է,ӶӰֵѡȡ.ˣԶ֮ǰ,ֱͼƶƽƽԤ

GiʾҶֵΪjƶƽֵ,giʾֱͼлҶֵΪiسֵƵ,nΪÿƶƽֵõĻҶȼ.ƶƽʵǶɢмƽŶ.һƶƽԲҪɶGi(nȡֵͬ)λƶƽ,.һ,жƶƽɴﵽЧ.ͼ2(c)Ƕͼ2(b)ƶƽĽ,ɼԭֱͼԸı,ŵ㱻.
,ͨԭʼͼֱͼƶƽ,׼ȷؼǰֵ,Ӷõɿֵзָ,ʹͷ۾Էֿ.ͼ2(d)ֵָĶֵͼB(x,y).
1.2۾λ
۾λһģƥ䷽1,2.Ȼ,һģƥӦСͬ,Yilleˣ6Xieˣ7ݽиĽĻڵģƥ沿ȡȻȶɿ,ԴżӦһȱ.һֳΪ۾λ,˷ֵָĶֵͼB(x,y),һЩ֪ʶ,Կٶ׼ȷضλ۾.,۾λĹҲ涨λĹ(ͼ3ʾ).۾λҪ裺
(1) ͨ1ɨֵͼB(x,y)λͷ.
ɨB(x,y)Ĺ,ٵ㷨ǳ1ֵ.һ˵,ڱΪ򵥵,Ϊ,һ鼴Ϊͷ,ͨǳ.ʹԹͷ׷˵,ͷҶҲǱȽС,ԶֵҲܽͷҳ.
(2) ͨͷĴ,λͷڱ߽,ͼI(x,y).
ڶλͷ,һδϵµľֲɨ,Եõͷ±߽.ͷпܳ0ֵߵ,ƽֵ.۲,Ƿ,߸һСֵ(޽ǵĲλǵͷ쵽·),ҳ.ڿһͼI(x,y),ͼΪͷ±߽ߵߵ,ΪҵСֵλ,׶Ϊ˼Ҷ˼1.5ͼֽ֮ĽСֵ.
(3) ͼI(x,y)Ӧ˹,2ɨ,һЩ֪ʶλ˫.
ʹI(x,y)еÿҶȳͼ3ʾľ

Fig.3The distorted differentia kernal
ͼ3εĶ΢ģ
ͼǱ߽е,ֵָͬ,Ȼȫͼɨ,붨λͷͬ,ҳ10,ֶԴ,ǳΪ˫۵ĳɶԿ,ĹбпܳΪ۾һ,ȻȡһԿڵĵΪ۾ĵ.
б۾õҪǣ
(a) ۾(width)߶(height)(area)Ӧ㣺
widthheight and area0.5;
(b) ۾鵽I(x,y)ϱ߽ľd1䵽I(x,y)±߽d2Ӧ㣺
d1d2(۾Ӧϰ벿);
(c) ۾ӦλͼI(x,y)ߵ.
۾λĽͼ4ʾ.

Fig.4 The examples of eyes location
ͼ4۾λʾ
1.3һ
۾ĵ̬йһͼI(x,y)ĵlתlﵽˮƽλ,ͼI(x,y)ȻI(x,y)ű任,ʹͼInorm(x,y)ĳߴΪ128192,СĹһ.һ˵,ͼInorm(x,y)Ķüëϱ߽,׶´(˶),˾ȧǸ(ͼ5ʾ),ʹĹؼ͹ֳ,СͼʶķΧ.


Fig.5 The examples of face normalization
ͼ5ͼһʾ
2 ʶ
ECVʶ㷨һֻͼȫƥ䷽.ڹһͼϽ,ȻǽиҶõؼȫֵ,ı,Щƥʶ.
2.1
ߵ˼ǻڻͶӰ.ͼĳڽлͶӰķʶھѱ,ڳȡֲ.Takeo Kanade ˣ8ô˷ȡֲ(졢ǵ),׼ȷʴﵽ75.,Brunelliˣ9ݶ˸Ľ,׼ȷߵ80.Ȼ,׼ȷδﵽĳ̶,ڲ˻ͶӰ˼Ļ϶˸Ľ,ȫֻͶӰ,,.
߿ɷΪ3,乹£
(1) ͼInorm(x,y)ֱСлͶӰ(ͼ6(a)ʾ),õc1c2.

(1)

Fig.6
ͼ6

(2)
(2) ͼΪԲ,qΪ뾶һԲ,Բ뾶Ϊpֵ,ÿֵͼΪԲ,Բĵ÷ֵľΪ뾶ͬԲ,Բÿ1ȡһ,360ĻҶ,õc3.Ȼ,XϵܵΪp(ͼ6(b)ʾ).gi()ʾiԲϦǶȴĻҶֵ,c3ʽ(3)ʾ.

(3)
(3) ͼΪԲ,qΪ뾶һԲ,Բ뾶Ϊpֵ,ÿֵͼΪԲ,Բĵ÷ֵľΪ뾶ͬԲ,Բÿ1ȡһ,ֱ180ȹ180180ȹ180㣬Ҷ,õc4c5,Ȼ,ЩXϵܵΪp(ͼ6(c)ʾ),(0ip,giͬʽ(3)).

(4)

(5)
2.2ȡ
ȫֵĳȡͨ5߽иҶõ.Ҷ߽һҪʾ,ڴʾ.ĳM,԰߿ڸƽ(ͼ7(a)ʾ),߶˵㿪ʼ,߸һͿɵõһ,еĸҶ任Ϊ߸Ҷ.

Fig.7Eigencurves
ͼ7
f(x)ʾ,NΪϵ,丵ҶӿɱʾΪ

ʵ,ǰ10ԸҶӼܺܺõرʾ.ͼ7(b),(c)ʾ,c1ԭ ,c1Ϊǰ10ӾҶ任ɵ.,ÿֻ 10ԸҶʾ.5ߵĸҶ,õһά Ϊ50ȫı.
2.3ʶ
ȫĳȡʹÿһ50άıӦ,ͨڷʶ.
V1,V2֮Ծȶ£

һȿԲ.֪ӦıX1,X2,...,X Lʾ,δ֪ıXʾ,ôɼX֪ľD(X,Xi),
D(X,Xi)D(X,Xj)(j=1,2,3,...,L,ji)
жi漴Ϊѡƥ.
3 ʵ
ʵ齨һݿFACEDB֮.FACEDB260.ÿ5Ųͬͼ,3ǲͬͼ,ǲ̬ͬ(ֱб)ͼ,ͼ8ʾ.,FACEDBй1 300ͼ,ЩͼֳཻļϣͲԼ.250˵ͼ(һѡȡƽ߽͡Ͼȵͼ),Ϊ֪ͼݿ⣻Լͼ,Ϊδ֪ͼ.


Fig.8 The face images in database
ͼ8ݿʾ
ʵֽУ(1) ϵͳޡܾµʶܣ(2) ϵͳ롰ܾµʶ.ܽ1,ϵͳƽʶʴﵽ95,ÿʶٶ3.ͬʱ,ʵ,ϵͳı仯̬ı仯нϺõ³,Թ³ϲ.
Table 1Test results
1ʵ

TestCorrect(%)Incorrect(%)RejectThreshold
Test 1 (no rejection)94.535.470
Test 2 (permit rejection)96.382.711.200.84

ʵ,ʶ,۴,ܾܾ,ֵ,޲1(޾ܾ),߲2(ܾ).
4 ܡ
ߵECVʶ𷽷ԭ,Чʵ,ѧϰ,ٶȿ졢׼ȷʸߵص.ʵ,˷ڲͬ顢̬ʶкܺõ³.ȱǣϴĹӰʶ.
,㷨չ Żܸõ,Կ˷Ӱ죻 ȡԱ,Ӧӵĳ.
оõ㽭ʡʽ(No.1997-7-15).¸,Ů1972,ʿ,ҪоΪý壬ݿ⣬ͼ,˹.꿵,1945,,ʿʦҪоΪѧϰ֪ʶȡרϵͳý/ý弼.Ҷ,1945,,ҪоΪͼ.
ͨѶϵ:¸, 310027㽭ѧѧ빤ϵ
ߵλ¸(㽭ѧѧ빤ϵݡ310027)
꿵(㽭ѧѧ빤ϵݡ310027)
Ҷ(㽭ѧѧ빤ϵݡ310027)
ο
1Chellappa R,Wilson C L,Sirohey S.Human and machine recognition of faces:a survey.Proceedings of the IEEE,1995,83(5):705740
2Samal A,Iyengar P A.Automatic recognition and analysis of human faces and facial expressions:a survey.Pattern Recognition,1992,25(1):6577
3Turk M,Pentland A.Eigenfaces for recognition.Journal of Cognitive Neuroscience,1991,3(1):7186
4Hong Z.Algebraic feature extraction of image for recognition.Pattern Recognition,1991,24(3):211108
5Narendra K S,Parthasarathy K.Identification and control of dynamic system using neural networks.IEEE Transactions on Neural Network,1990,1(1):427
6Yuille A L,Conen D S,Halinan P W.Feature extraction from faces using deformable templates.In:Harvard Robotics Lab.ed.Proceedings of the CVPR'89 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Washington,D C:IEEE Computer Document Press,1989.104109
7Xie X,Sudhakar R,Zhuang H.On improving eye feature extraction using deformable templates.Pattern Recognition,1994,27(6):791799
8Kanade T.Picture processing by computer complex and recognition of human faces.Technical Report,Department of Information Science,Kyoto University,1973
9Brunelli R,Poggio T.Face recognition:features versus templates.IEEE Transactions on Pattern Analysis Machine Intelligence,1993,15(4):10421052
1998-10-22յԭ,1999-04-06յ޸ĸ
