о뷢չ
JOURNAL OF COMPUTER RESEARCH AND DEVELOPMENT
1999Vol.36No.7P.876-881



ڱĵģƥ䷨
־¡
ժҪڵģƥ䷨Ļϣ˻ڱĵģƥ䷨ͼʶ.÷KL任ģĶλʹöλٶȶߣʶ׼ȷҲõ.
ؼʣͼʶ𣬱ģƥ䣬Gabor任KL任
ţTP391.4
ELASTIC MODEL MATCHING BASED ON EIGEN-FEATURE
XIONG Zhi-Yong 
(Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100080)
SHEN Li 
(Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100080)
AbstractThis paper brings forward a new elastic model matching method based on eigen-feature to recognize facial images. The method utilizes KL transformation to locate the model so that the accuracy of model locating is improved and the procedure is also accelerated. The recognition ratio is improved too.
Key wordsfacial image recognition,eigen-feature,elastic model matching,gabor transformation, KL transformation, refining step by step
1
˹о򣬶ʶоģһֱоص㣬ͼʶǵǰһоȵ.ΪĶ䣬仯Լڸָı仯߶ȡաתȣʹͼĻʶкܴѶȣҲսԣӶҲ˶ʶ𷽷ģʶ𷽷1ݣʶ𷽷2ݣɷַı(eigenface)34ݣGabor任ĵģƥ(elastic model matching)513ݵ.
ÿַŵ㣬Ҳв6.ķ洢ռҪСʶٶȿ죬ʶʵͣ緽ҪѵȽϳֲܵõϺõʶЧĿոı䡢߶ȸıԼСαܺܺõʶ𣻱ʶٶȿһŵ㣬Ҷͼʶ׼ȷʺܸߣĿոı䡢߶ȱ仯ԼλƱ仯ʱܺܺõʶ6ݣģƥ䷨Gaborͼб任õı任ϵΪʶ𣬲õƥ䷨Դﵽģľƥ䣬ӶܹСαĸţGabor任оֲԡͨԼѡԣܹܺõȡĿı7ݣĿֲ任СתԼոıܵõϺõʶ5ݣ˸÷һֽķ.һȱݣʶٶ.Ϊõʶǽģ徫ȷƥĻϵģĸԣͬʱʵԣ6.ܹģ徫ȷƥľٶȣĸԣɽٽʵ.
Ϊˣڵģƥ䷨Ļϣһµʶ𷽷ͬGabor任ȡĿγģ壬ʹKL(Karhunen-Loeve)任ģĶλʹöλٶ뾫ȶ.ʵʶЧҲõ.ĵ2ּҪܵģƥ䷨䶨λ̵ʱ俪3ֽܻڱģƥ䷨ıĲңӶģĶλٶȣ4ָһЩʵǽչ.
2ģƥ䷨
2.1
ģƥ䷨ôͼ(labeling graph)ʾһͼͼʶתΪеģͼ(model graph)ʶͼ(image graph)֮ƥ5.ͼ(image graph)Զλ㷨Ĺؼλȸʶʾ͸ߣλʶһͼҲ죬ʵó̶Ⱦ͸.
2.2λʱ临Է
ף5Уʹɶλȫֶλֲλ.õĴͼΪһε󣬹70㣬ͼ1(b)ʾ.ֱʾʽѡûг㣬۾ӡ첿ȣƥ侫Ȳߣ˲ͼ1(c)ʾ(ͼ31).ѡȡҲɲοף13.


ͼ1(a)ͼ(b)ε(c)(ģ)
ͼάΪNN󳤶ΪlΪwڵΪmΪʵȫֶλҪ(N-l+1)(N-w+1)mαȽ.ÿαȽϵʱ俪Ϊtcȫֶλʱ俪ΪTg=(N-l+1)(N-w+1)mtc.ȫֶλһԵĶλ̣Ϊ˵õÿڵĸȷλãͨÿڵ㻹Ҫоֲƶ5.ƽÿһڵҪƶȽϲȷλ(ʵΪ248)ֲλʱ俪ΪTl=mtc.λʱ俪ΪTl=tg+tl.
ʵζλʱ俪ܴʶٶԭ֮һҾζλ󣬶λҲܾȷ(ﶨλ׼ȷָÿһ㣬λƫ5ؾ)ƥʱڿеÿһģ壬ȻٽͼȫֶλҪӦľֲλйPģ壬ΪʶһʶͼܵĶλʱ俪ΪT2=Tg+PTl.
Ϊ߶λȷȼٶȣԿǴиĽһַǼٵĽڵӶȽϴܹ٣ʵ⽫¶λ׼ȷ½ӶʶҲ½һַWiskottĵԴģƥ似(elastic bunch matching)3ݣʹôṹͼ(bunch graph)ΪͨģıʾͼĶλʹöλ׼ȷȴߣӶƥʱڿеĸģ岻ٽȫֲĶλֱƥ䣬ʹʱ俪Ϊ.ַһȱݣʹôṹͼͼȫֶλʱڴṹͼÿڵ㴮Ҫѡһѵƥӽڵ(expert point).贮ṹͼmڵ㴮ÿڵ㴮kӽڵ㣬ΪʵȫֶλҪ(N-l+1)(N-w+1)mkαȽϣȽϴˣȫֶλʱ俪Ҳ֮ӣϾֲλʱ俪ṹͼĶλΪT3=kTg+TlȽT1T3λʱ临.ƥʱٽпģĶλܵĶλ䣬ΪT4=kTg+TlPkȽT4T2ĽǺԵ.
ɼܽһṹͼĶλ٣λƥʱ俪Ҳ֮.ˣܹҵʹģ嶨λ׼ȷķģƥ䷨ʶٶȼʶ׼ȷʶߣʵ.Ϊˣ˻ڱĵģƥ䷨.
3ڱĵģƥ䷨
ɼģƥ䷨ģĶλҪܸߣĿҲΪʹģбʾͫסǼ⡢򤡢ȵĽڵʶͼӦĽڵбȽϣƫʧģ嶨λתΪֲ̽⣬ĳЩؼλȷˣʶͼλҲȷˣף5ݡ13еʶ㷨ͼʶ.
ֲ̽⣬Ѿ෽ԵͶӰ10ݣ֪11ݵȷ룬￼ʹKL任ֲ̽.
3.1KL任
KL任Ҳɷַ(PCA)һͼıʾ89Լʶ34.ѵͼеKL任ɵõ(eigenfaces)ռ䦣faceKL任ľʵϸڿɲοף3489.
3.2󾫶λ
Ϊ˽ֲ̽⣬ȫ󾫶λͼΧڲ۾򡢱첿ȣҲԲ󾫶λȴһͼнҳȻΪ۾򡢱첿Ⱦֲȷ̽ͬʱɵõͫףǼ⣬촽λ.ЩͿγһֲ󣻸λöӦϵɽڵλƲٽв.
ʹ󾫶λֲĿռλùϵС̽ⷶΧʹöλҲõ.ˣʹ󾫶λ.
3.2.1̽붨λ
ͼ2(a)ɼڴʶͼռȡһСľO()faceռͶӰعȲõع()ٵһλȡһͬfaceռͶӰعõعͼռ䣬ȡعСľΪ.罫ĵ趨ΪǼ⣬򸽴ؾ˱ǼĶλ.
3.2.2ֲĶλ
3.1ڿ֪ȡĸֲΪѵѵͬγ۾ռ䦣eyeӿռ䦣noseռ䦣mouthȾֲռ䣻ΪȡӦСľӦľֲռͶӰõӦع綼ȡعСľΪʶ򣬾ܹɸֲ̽.ͬʱ趨ĵΪӦ㣬ɶͫףǼ⣬촽λãͼ3ʾ.
3.3λʱ临Է
֪󾫶λҲ.Ϊ2.2бȽϣ3.2.1Уͼά䣬O()ĳΪlΪwΪʵĶλҪ(N-l+1)(N-w+1)αȽϣÿαȽϵʱ俪Ϊtfaceλʱ俪ΪTf=(N-lface+1)(N-wface+1)tfaceռΪķ֮һ򣬿ɽΪ䶨λʱ俪ΪTf/4ͬ첿򶼿Ϊ䶨λʱ俪ΪTf/4ܵĶλʱ俪ΪT5=7Tf/4.ɽΪtfacetcȣT57Tg/(4m)ƥʱڿеÿһģ嶼ٽͼĶλֱƥ伴ɣܵĶλʱ俪ΪT6=T5.
ȽT5T1T3,T6T2T4Է֣ʹ󾫶λڵģĶλʱ俪٣ͬʱڿеģƥ䣬λʱ俪Ҳ٣ʵλ׼ȷҲܸ(οʵ12).
3.4Ч
3.4.1Գ߶ȵĴ
ף3ݡ4ݿ֪PCAͼ߶ȱ仯Ӱܴ.ף4нͬ߶ȵͼ񼯻ѵõͼ񼯵ıռ䣬ʶЧ̫.￼ǽͬ߶ȵͼ񼯷ֱѵõͬ߶ȵͼռĸ.ֲ̽ʱȷֱòͬ߶ȵıֳ߶ȵıȷʶͼĳ߶ȣȻӦ߶ȵ̽ʶͼĸֲ㣬õʶͼĵʶʱȶԴʶͼGabor任ȡϵֵȻõģƥ䷨ʶ𣬹ڲͬ߶ͼʶºܶ.һ˽ɲοף4ݡ14.
3.4.2תĴ
תͼʶͬҲɽͬתǶȵͼ񼯷ֱѵõͬתǶȵͼռĸ.̽ʶԲͬ߶ͼĴ.
4ʵо
4.1ʵ
ʵõͼԴʡѧԺýʵң12.Ϊ3ֹ¶16ͬ˽㣬ѵѡµͼ񣬹16.ԼѡҲԴͼ32.
4.2ʵ
ʵ1. Ĳ붨λ(3.2.1).
ñ(eigen faces)ڴʶͼҳ.ʵͼ2ʾ.ʵ̽׼ȷʴﵽ100%.ʵ7.


ͼ2(a) ѵͼ񣬾οΪֹѡȡ
(b) ͼ񣬾οΪ̽⵽(ĿӰ֪ղͬ).
ʵ2. ֲĲ붨λ.
ֱõģƥ䷨ȫ󾫶λ󾫶λֲĲ붨λ.(עͬģƥ䷨ԽĶλһҲʹø÷ֲĲ붨λ.)ȫ󾫶λ󾫶λУ7.ʵ1ʾ.ͼ3ʾΪ󾫶λֲĶλ.
1ʵ2

λ׼ȷʵģƥ䷨ȫ󾫶λ󾫶λ
44%67.7%100%
56%33.3%93.3%
50%33.3%100%


ͼ3(a) ѵͼ񣬾οΪֹѡȡ
(b) ͼʮֽߵеΪ̽⵽㣬ҲǾֲĵ(ĿӰ֪ղͬ).
ʵ3. ͼʶ.
ֱñ(eigen faces)ģƥ䷨Լڱĵģƥ䷨ͼʶ.ʶ2ʾ.
2ʵ3

ʵ鷽
ģƥ䷨
ڱĵ
ģƥ䷨
ʶ63.3%90%96.7%


ʵ4. ڱĵģƥ䷨뵯ģƥ䷨ʶٶȱȽ.
MatlabԱдƵ200MHz΢.ֹʱֻڱĵģƥ䷨ȵģƥ䷨ʶٶȴԼ20%.
4.3ʵС
ʵ1ɼKL任ĲҺͶλ׼ȷʺܸߣʵ2ģƥ䷨ھֲĶλ׼ȷʲߣͬȫ󾫶λھֲĶλ׼ȷҲߣܹӰ.󾫶λھֲĶλ׼ȷʺܸߣﵽЧʵ3ڹոıʶʽϵͣҲһ֤÷ܹӰ.ģƥ䷨ܹӰСʶʽϸߣڱĵģƥ䷨֮ģƥ䷨ʶ𾫶ƣʵ4ڱĵģƥ䷨ʶٶҲ. 
5չ
ڵģƥ䷨Ļϣ˻ڱĵģƥ䷨KL任ģĶλʹöλٶȶߣʶ׼ȷҲõ.ҪĹǽһƥٶȣʹ㷨ʵ.
ĿõйѧԺ塱оĿ(ĿKJ952-J1-707)ʡ.
߼飺־£У19719ʿооΪѹģʽʶͼ񡡡.
У193711ڣʿʦҪоΪ㡢ģߡ硢Ŵ㷨ģϵͳݴ.
ߵλ־(йѧԺ㼼о100080)
(йѧԺ㼼о100080)
οף
1Brunelli R, Poggio T. Face recognition: features versus templates. IEEE Trans on PAMI, 1993, 15(10): 10421052
2Cottrell G W, Fleming M. Face recognition using unsupervised feature extraction. In: Proc Int Neural Network Conf, vol 1 Paris, 1990. 322325
3Turk M A, Pentland A P. Face recognition using eigenfaces. In: Proceeding IEEE Computer Society Conf of Computer Vision and Pattern Recognition, Hawaii, 1991. 586591
4Turk M A, Pentland A. Eigenfaces for recogntion. Journal of Cognitive Neuroscience, 1991, 3(1): 7186
5Lades M, Vorbruggen J, Buhmann J et al. Distortion invariant object recognition in the dynamic link architecture. IEEE Trans Comput, 1993, 42(3): 300311
6Jun Zhang, Yong Yan, Martin Lades. Face recognition: eigenface, elastic matching, and neural nets. Proceedings of IEEE, 1997, 85(9): 14221435
7Field D J. What is the goal of sensory coding. Neural Computation, 1994, 6(4): 559601
8Sirovich T J, Kirby M. Low-dimensional procedure for the characterization of human faces. Journal of the Optical Society of America (Series A), 1987, 4(3): 519524
9Kirby M, Sirovich T J. Application of the Karhunen-Loeve procedure for the characterization of human faces. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990, 12(1): 103108
10Kanade T. Picture processing by computer complex and recognition of human faces. Kyoto University, Tech Rep: 1973
11Vincent J M, Waite J B, Myers D J. Automatic location of visual features by a system of multilayered perceptrons. IEE Proceedings, 1992, 139(6): 405412
12ftp://ftp.media.mit.edu
13Wiskott L, Fellous J M, Krger N, von der Malsburg. Face recognition by elastic bunch graph matching. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1997c, 19(7): 775779
14Buhmann J, Lades M C, von der Malsburg. Size and distortion invariant object recognition by hierarchical graph matching. In: Proc IJCNN Int Conf Neural Networks, San Diego, CA, 1990. 411416
ոڣ1998-11-16
޸ڣ1999-03-02
