Զѧ
ACTA AUTOMATICA SINICA
1998 24 4 Vol.24 No.4 1998




һֻģӽṹдʶ𷽷1)
Ф졡Ϊ
ժҪṹƥһЧддʶ𷽷Ϊ˼ƥ㣬һֱѰò ƥĽƥķ.ƥṹƥۺϵĻϣ3.755һ 꺺ȡ45ӽṹṹƥ.Щӽṹܳ׻βǵļȽ.ͬʱͨӽṹģ弰ӽṹ̬ȡ㷨ʹӽṹƥ׼ȷȵõܴ.ʵṹ÷ʹṹƥԼ50%Ƶʶһ.
ؼʡдʶ𣬽ṹƥ䣬ӽṹģ.
ONLINE HANDWRITTEN CHINESE CHARACTER RECOGNITION
GUIDED BY COMPONENTS WITH DYNAMIC TEMPLATES
XIAO XUHONGDAI RUWEI
(Institute of Automation,The Chinese Academy of Sciences,Beijing100080)
AbstractStructural matching is a main approach for online Chinese character recognition.In order to reduce its great computational comple xity and improve its performance,people have been seeking for a way to guide the whole matching by the result of partial matching.In this paper,the authors prop osed 45 basic components from 3,755 categories of the dailyused Chinese charac ters to guide the stroke segment matching.Because they always locate at either the beginning or the end of the stroke segment string,these components are easy to extract and separate from other parts of the character.Besides,the reference templates of these components are dynamically extracted from the reference segmen tstring and dependent on the current matched character so that a more accurate matching is carried out.Experiments show that the segment matching computation h as reduced almost 50%.The approach is also enlightening for other similar object matching problem.
Key wordsOnline bandwritten character recognition,structural matching,substructure,dynamic template.
1
űЯʽʼǱԵķչӿܵߣдʶΪһֿ ļдʶĺԶ¼뷽ʽһ˹оߵȤ.ڶԹʶȡ һɼ֮ǰоصתƵ˳仯ʵдֵʶ. ȻеʶϵͳϺõؽ˴ʵĺֵʶ1ݣһЩϵͳϺõ ˶ֱ˳ƣ2ݣûһϵͳͬʱ.
дһ㶼Աʻеķʽ洢һڱʻʶδ ƥĽṹƥ㷨дʶеõ˹㷺ӦãͨԸṹԪӼԣṹһ̶պֵα䣬㸴ӶȽϸߣԱ˳ֽʶ.˽ѻʶķʶ֣תΪάͼƥ䷽ʶ.ƥ䷽ŵģ弰ƥ㷨ʵֺ.ڲʵ£˳ı仯Ӱ캺ֵͼƥ䷽ʶûʵд֣Ч˳ʶӰ죬ʱʻӵıʶεºͼı仯ӶʶӰ죬ͬʱĿǰƥ㷨Ժα൱ޣǶдϰߵԭֵαѻִö࣬ƥķʶǲ. Ϊһ͵ģʽʶ⣬ʶķչڶԺֽṹʶֹģˮƽ.
дϰߵԭں຺ܶжָӽṹ ļʻд.磬д֡ʱдƫԡڡд֣ںֵ֡ддƫԡ֡.ڴˣܵϵͳƥ䷨ֽдַ࣬Ȼ׻βӽṹԴƥɺʶϸ.Щӽṹܳ׻βıʻУֻҪһʵƥ㷨ǵļǲѵģ⵽ʶĺ ĳһӽṹʱֻаӽṹĺѡֲʸʶֽнṹƥ䣬ؽ˽ṹƥ㣬ͬʱҲʻ˳仯µʶ.ڱĵĺУֱʶϵͳģ͡ӽṹĹ켰ƥ˽.
2оϵͳģ
ʶĽչǶԺֽṹأԺֽṹֵܵд ǶԺֵĸ̵֪Ӱ죬ӶһͨĺƵʶ𷽷ͨʻĹϵʶֲ׻ӽṹͨЩʶӽṹĹϵֵʶ.˼룬ڰʮĩӭȾһʽģǿ ƥʶ𷽷3.ַӰȲͬ˳еıʶзָӦģƥ.ʲôʱĴѰָʲôʱ ĴѰָ?һ൱ѵľߣûһ㷨ܶһ̽ѧ ϰȷľߣʹֹͨÿһֵĽṹ˳ְ֡ ؽ̸дֱʻ̶ʶηָ㷨ƵԭҲ޷ȷ򴮵һλȥѰӦָʶ𷽷ʵ൱ѵ.
Ϊ˳ֵдֵĶ̬̬Ϣһͼƥַ ࣬ṹƥ䷨ϸдʶϵͳ4ݣַΪÿʶṩɺѡϸ̽ʶѡֵԱʶδнṹƥ䣬ۺƥṹƥĽõʶֵ.ṹƥǸʶϵͳҪڣִԭҪȣһɺܶʻɣһʻλñ仯пܵ±ʶεԣλáıʻȵı仯ֱʻԽ࣬ԲȶԽࣻΣֱʻһ࣬ƥ䷽ľԣṹľֲпܱڶ֮𣬲֣硰Ρ롰֣Ǳ˳仯һ̶Ӱṹȷƥ.ṹƥ䷨һҪȱ㸴Ӷ̫.ܸݺֵĲֽṹȷǷֵĽṹƥ䣬Ϳɴؼƥ㣻ͬʱѺַֽɶÿһĽṹԼ򵥣ҲԼƥ.
ںֱʻʼĩβںܶṹȶĲ׻ָλõԣȽ װǴӱʻз.ˣĶ45ӽṹͨӽṹṹƥ.ϵͳͼͼ1ʾ.

ͼ1ʶϵͳͼ
3ģӽṹĹ
ʶӽṹѡҪѭ¹
1)ЩӽṹںֵĿʼĩβıʻ.ںʼʻеӽṹΪ ӽṹںֹʻеӽṹΪβӽṹ.
2)Щӽṹнǿķ.һӽṹķҪȡأ ӽṹʻںѡгֵƵ.ṩĴʶֵкѡֶĳӽṹһֵƥûκãΪеĺѡʶֽƥ䣻һ棬ĳһӽṹڲĴʶֵĺѡַгԽ࣬ҪʶֽƥĺѡԽ.
3)ӽṹĿ.ѡӽṹ̫٣򼸺кѡʶֽ ƥ䣻ѡӽṹ̫࣬ӽṹϵʱҸӽṹ׻Ӷʶϵͳܽ.
4)ͬ˶Щӽṹд˳ȶƥ㷨ľޣпܵӽ ƥʧܣӶʶ.
ͨʵ飬ѡ45ӽṹ3755гֵƵ2106. 1иֱΪ45ӽṹơż3755һгֵƵ.
1ӽṹơżƵ

ƵƵ Ƶ
05015103053
122162431ɽ25
2ҳ3017173233
32151825337
421319ʯ4834ʬ32
5ľ13320423544
64421253635
74222433723
87723753811
92024Ŀ263914
1013257140199
112226514122
121327Ů5242ڥ86
131228284344
1431292044Ѩ19


Ϊ0-2ӽṹΪβӽṹΪӽṹ.ӱпԿѡ ӽṹ顱ࡱȳƫԣҪΪƫԡед˳ܲȶеϰдƫԡдֵһЩ෴ڲѡƫԡ顱ࡱҪΪЩƫĲ׼ʻ.
3.1ӽṹԵȷ
ӽṹӽṹҪ֮һݴˣǶÿһָӦ.һ ְĳӽṹʱӽṹűΪúֵӽṹԣ򣬸úֵӽṹԱָΪ1.ĳְһϵӽṹѡŽСӽṹΪ.纺֡ȡȰӽṹСְᡱڡᡱСڡС ԱΪӽṹᡱıţΪ1.ӽṹŵĵڶþǷӳӽṹƥ˳ʶֵĺѡּṩӽṹӾСŵӽṹʼƥ.ΪΪӽṹʱͿǵӽṹĲְƹϵһְӽṹ.ŽСӽṹпܰŴӽṹӽṹաаӽṹڡ.һְӽṹʱһ˵бŽСӽṹнСĳƵǿӦƥ.
3.2ӽṹģ
ӽṹʶ𷽷Ҫ⣺һӽṹĳȡǶͬһ͵ӽṹ 䴦ڲͬеĲͬλʱСʻıλùϵҲӦر仯Ӷƥʶ.Ϊ˽һ⣬ķ˲ӽṹβӽṹ⣬ñʻ̶дֽģ弰̬ƥ̬ȡʶֵӽṹ.Ϊ˽ڶ⣬˻ģĸ.ǲΪÿһӽṹɱʻɵĹ̶ģ壬ֻ¼ÿһӽṹıʻĿƥʱٶ̬شӺѡֵıʻģʼֹλóȡͬʻɵıʻΪӽṹӦڸֵĻģ壬ںѡֵıʻģǴӱʻ̶ѧϰõģ4ݣÿһӽṹıʻĿҲǹ̶ģӶ׶׼ȷصõӽṹģ.ģĽʹǴΪÿһӽṹܶͬλõģаѳӶʹӽṹֵƥΪ.
4ӽṹĽṹƥ
ַ㷨Ϊʶṩĵiѡwj(i=1,2,M)ʾѡwjдڵжӽṹ(ӽṹԲΪ1)ӽṹaiʾӽṹĽṹƥͼ2ʾӺСǸӽṹԵĺѡֿʼ(ֵúѡֲӽṹ)Ӹúѡı׼ģʶгȡӽṹƥ䣬ƥСth֤ӽṹɹƥ̬ȷʶӽṹֹ(ӽṹ)ʼ(βӽṹ)ʶελãӱ׼ģ弰ʶɾȥӽṹıʶΣʣĲƥ䣻ӽṹƥ䲻ɹеӽṹƥcith򽫴ʶѡֵı׼ģһƥ.


ͼ2ӽṹĽṹƥ
4.1ӽṹĳȡʶ
ڲοģй̶Ŀıʻ׻βȡӽṹǲ ʶͬڿܴڵʣʹӽṹĿСڻڱ׼ģıʻˣӽṹĳȡοģֱӣΪˣķԴʶӽṹ˶̬ĳȡ.
ĳȡӽṹı׼οģnʻɣԴʶֵǰnʻ(ӽṹ)nʻ(βӽṹ)ɵıʻΪӽṹĳʼʻģ壬ȻÿһʻֽɱʶΣ5.Ϊʶʵĺ֣ÿһʻյڵһʻʼʻ߶αΪʶ.ʶ˳Уγɳʼʶδ.Ȼѡӽṹʱ˳ȶΪ˱żȻı˳仯ӽṹԱʶ򣬸ʶεȺ˳Ҫԭȷ.
1)ʶμԵ»ҵĹϵʶΰ»ҵ˳ .ͼ3Уʶl1ʶl4Ϊҹϵʶl8ʶl6Ϊ¹ϵʶl1Ӧڱʶl4ǰ棬ʶl8λڱʶl6֮ǰ.


ͼ3֡աʶηֽ
2)ʶμ䲻Ե¡ҵĹϵΪ¡Ϲϵʶΰд Ⱥ˳.ͼ3бʶl4ʶl8Ϊ*ϵ(*ʾϵȷ)д˳򣬱ʶl4Ӧڱʶl8֮ǰ.
Ϲ򣬵óͼ3ӽṹաıʶδΪl1l2l3l4l8l7l5l6.ӦָǣֻܶԲֱ˳еӽṹƥܵ˵ʺڹ̶˳ıʻ.
4.2ʣಿĳȡ
ʣಿָʶֱʶδѡֵĲοģʶδȥӽṹʶδʣµıʶ .ڲοģӽṹй̶ĿıʻȥӽṹıʶζõʣಿʣಿĳȡҪԴʶֶ.
ʵĴڣݲοģṩıʻȡʼʶδʱӽṹĩβ βӽṹĿʼпܰһЩʣಿıʶ.ڴʶֳʼӽṹʶδӦӽṹĲοģʶδƥɺӽṹʶδʣಿɸ¹̬ȷ
1)ӽṹĳʼʶδݱʶдȺ˳Уӱʶδĩ˿ʼ ֱܴӲοģӽṹҵӦʶεıʶΪֹ.ڸñʶ֮ǰ(ñʶ)ӱʶδΪӽṹʶδıʶʣಿ.
2)βӽṹĳʼʶδݱʶдȺ˳УӱʶδĿʼβ ֱܴӲοģӽṹҵӦʶεıʶΪֹ.ڸñʶ֮(ñʶ)ıʶӽṹʶδǰıʶʣಿ.
4.3ṹƥ
ÿһʶΣӦľԼ.԰ʶ벿(ӽṹʣ ƥʱ)ꡢʶγȼ˱ʷȣʶε԰ʶ˳򡢸ıʶδȣ4.ӽṹƥĽֱӰֵƥ䣬ӽṹиʶε˳ȶˣȶӽṹʶнȻͨзѣϲĶ̬滮ƥ1ݣӽṹƥʣಿ֣ʶθд˳гɱʶδͨ׼ɳ۷ƥ6.
5ʵ
ֿеһ3755ú֣35׹ϵͳѵ ׼ģ壬Ȼ3еȹ̶(вʣ˳仯дб)һвԣַ㷨Ϊÿʶṩ10ѡΪԱӽṹĽṹƥ㷨ܣ²ԣ
1)Ϊ˼ģϵͳܵӰ죬ʵ.ȣڶÿ຺ֵӽṹ Խ˹ѵ(ֵֹ)󣬴ӾͬӽṹԵĺּԶѧϰһӦӽṹı׼ģ壬Щ׼ģԴʶֽӽṹ⣬ȷʽΪ86%ûģӽṹ⣬ȷʴ98%.
2)ñķƽÿʶ10ѡе5.1ƥ䣬ƥ Ȼȫѡƥ䣬ƥԼ50%.
3)ƥϵͳʶΪ93.9%ӽṹƥʶʻ. ҪΪӽṹƥ㷨ȻԺα估˳仯ǿˣӽṹƥ䱾ȷʲ100%ۻڼǰ10ѡۻʶʾͽ97%ʶʾкܴ.
6
ƥṹƥۺϵĻϣһӽṹĽṹƥʶ𷽷.ͨ ʵӽṹβӽṹӽṹĽֱʶδƥ䣬˽ṹƥ㣬ͨΪӽṹģ弰ӽṹ̬ȡ㷨ʹӽṹƥ׼ȷȵõܴ.дȡ˺ܴɾ͵Ľ죬ʶ𷽷ͻڶģˮƽӽṹļ⼰ӽṹϵʶǶʶ̵ֹһģ⣬ĵķԻӽṹʶ𷽷Ǻ.
1)ȻѧĿ.
߼飺Ф졡1992ҵйƼѧߵѧϵ1997йѧԺԶʿѧλ.ҪоΪģʽʶ˹ܡͼ.
ΪҪ˹ܡģʽʶо19801982նɴѧϵѧߣпԺԶоԱʿʦпԺԺʿģʽʶ˹ܡ࣬廪ѧͷѧȽʮУļְ.
ߵλйѧԺԶо100080
ο
1Tsay Y TTsai W H.Attributed string matching by splitandmerge for online Chinese character recognition.IEEE Trans.on Pattern Analysis and Machine Intelligence,1993,15(2)180-185
2ӭΪ.дʶνṹ.Զѧ198814 (3):10-17
3ӭΪ.дʶʵ.Ϣѧ19892(4)18-30
4Xiao Xuhong.Online handwritten Chinese character recognition by merging ima ge features with Dynamic information.the 2nd ChinaKorea Joint Symposium on Ori ental Language processing,1996,65-70
5Kasturi,Raman et al.Document image analysis:an overview of techniques for gra phics recognition.preproceedings of int.Workshop on Syntactic & Structural Pat tern recognition,1990,192230
6Xiao Xuhong,Dai Ruwei.Online Chinese signature verification by matching dyn amic and structural features with a quasirelaxation approach.in:Proc. 5th Inter national Workshop on Frontiers in Handwriting Recognition,Colcheetel England 1996
ոڡ1996-09-18
