·512· 北京科技大学学报 2005年第4期 一个亟待解决的问题， [4)金连文，彭秀兰，尹俊勋，一种手写体汉字特征提取新方 法一小波变换及弹性网格技术的应用。中国图像图形 参考文献 学报，1998,7：549 [5]Suyken JA K,Lukas L,van Dooren P,et al.Least squares support [1]Hildebrandt T,Liu W.Optical recognition of handwritten Chin- vector machine classifiers:a large scale algorithm.Eur Conf ese characters:advances since 1980.Pattern Recognit,1993,26 Circuit Theory Des,1999,8:839 (2:205 [6]Hsu C W,Lin C J.A comparison on methods for multi-class [2]Suykens J A K,Vandewalle J.Least squares support vector ma- support vector machines.IEEE Trans Neural Networks,2002. chine classifiers.Neural Process Lett,1999,9(3):293 13:415 [3]Khotanzad A.Invariant image recognition by Zemike moments. [7]Kok SC.Efficient computations for large least square support ve- IEEE Trans Pattern Anal Mach Intell,1990,12(5):489 ctor machine classifiers.Pattern Recognit Lett,2003,24:75 Off-line handwritten Chinese character recognition based on fusion features and LS-SVM GAO Yanyu,YANG Yang",CHEN Fep 1)Information Engineering School,University of Science and Technology Beijing,Beijing 100083,China 2)JF Computer System Co.Ltd.,Beijing 100083,China ABSTRACT The proposed off-line handwritten Chinese character recognition system was composed of a feature extraction module and a recognition module.In the feature extraction module,the orthogonal Zernike moments and the elastic mesh technique were combined to get fusion features,which present the global and local features of hand- written Chinese characters and have great discriminative capability.As for the classification module,one approach that is very similar to the neural network classification strategy was used with the Least Square Vector Machine(LS- SVM),which not only has the excellent performance of generalization and recognition accuracy,but also can solve the multi-classification issue effectively.Experimental results indicated that the proposed method could get good recognition results. KEY WORDS off-line handwritten Chinese character recognition;least square support vector machine(LS- SVM);Zernike moment;elastic mesh一 5 1 2 - 北 京 科 技 大 学 学 报 2 0 5 年 第 4 期 一个 鱼待 解 决 的 问题 . 考 文 献 [2 ] [ 3 ] H il de b tna d t ,T L i u .W O P ti e al 传 e o咧 ti on o f h助 d 认理 i t e n C h i-n e s e e bar a e t e sr 旧 d v an e e s s icn e 19 8 0 . P a伽几 R eC o gn it , 19 9 3 , 2 6 (2 ) : 2 0 5 S u y k即 5 J A K , V 沁l d e aw 】l e J . L e as t s q u 别re s s u P ort v e c 加r m a￾ch 政 c las s iif ers . N e u ar l P or e ” s 玩t , 19 9 9 , 9( 3 ) : 2 9 3 hK 。 加口 . d A . nI v 硕ant 面ag e r ce o g n lit o n by Z冶m 溉 m o m e n ts . I E E E I 丫a . s P a t e 口 A . a l M a e h I o t山 , 19 9 0 , 12 ( 5 ) : 4 8 9 4[ ] 金连 文 , 彭 秀兰 , 尹俊 勋一 种手 写体 汉字特 征提 取新 方 法— 小波 变 换及弹 性 网格技术 的应用 . 中国圈 像圈形 学报 , 1 9 9 8 , 7 : 5 4 9 [ 5 ] S ly k. n J A K , L u k as L , v an D o ore n P, et al . L e ast s q u即 旧 5 5叩 p o rt v e cot r m a e h i n e e las s i 6 e sr : a l叱 e s e a l e al g ior ht m . E u r C o ilf C i代 u it T h eo 叮 D es , 19 9 9 , 8 : 8 3 9 [ 6 ] SH u C W, L in C J . A e o m Piar s on o n m het o ds for m u l ti 一 e las s s l l P P o rt v e c t o r m朋 h访 se . IE E E J n a n , Ne u ar l服wt o r ks , 2 0 0 2 , 13 : 4 15 【7 ] K ok 5 C . Eif c i ent e o m P u at ti o ns for l雌 e l e ast s q u a了e s u Ppo rt ve - e t or m a e 肠的e o las s伍 esr . P a t e 几 R eC o邵it L et , 20 0 3 , 2 4 : 7 5 参1[] O -f li n e h an d w r it e n C h i n e s e e h ar a e t e r re e o g n it i o n b a s e d o n fu s i o n fe a tu r e s an d L S 一 S V M 6 月口 aY ny “ ,气YA N G aY 心气C 月万N eF 尹 I ) I n fo n n at i o n nE g ine e r m g S e ho o l , U n ive rs ity o f s e i e cn e an d eT e hn o 1 0 幻 . B iej in g , B iej in g l 0 0 8 3 , Ch in a 2 ) FJ C呱P u ter sy s t em C o . L t d , B iej ing 10 0 0 8 3 , Ch in a A B S T R A C T hT e P r o P o s e d o -f l ine h an d w it t e n C ih n e s e e h a r a c t e r re e o gn i it o n sy s te m w as e om P o s e d o f a fe a h止 e e x tr act ion m o d u l e a n d a re e o g n l t ion m o du l e . nI hte fe a t u r e e x tr a o t i o n m o d u l e , het o ir 五o go aln Z e而ke m o m e nt s an d ht e e l as it e m e hs et e hn i q u e w er e e o m b i n e d ot ge t fus ion fe a 仅叮e s , w ih e h Per s e in ht e g l o b al an d l o e a l fe at ur e s o f h an d - w r it e n C h i n e s e e h ar a c t e sr an d h va e gr e at id s e ir m i n a t i v e e aP ab iliyt . sA for het e las s i if e iat o n m o d u l e , o n e aP P r o a c h ht at 1 5 v e yr s im i l ar t o ht e n e址a l n e wt o r k e l as s iif e iat on str at e gy w as us e d 侧t h het L e ast s q切叮e Ve 以o r M a e ih n e (L S - S V M ) , w h ihc n o t on l y h as ht e ex e e ll ent P e r fo n n an c e o f g e en alr i z iat o n an d er c o gn i t i o n ac c ur a c y , b ut al s o e an s o l v e ht e m u l ti 一e las s iif e iat o n i s su e e fe ict v e l醉 E x P e ir m e n ta l er s ul st in id e at e d t h a t ht e P r 0 P o s e d m e th o d e o u ld get g o o d er e o gn iti o n er s ul st . K E Y W O R D S o 作l ien h an d w it et n C h in e s e e h 别旧 e etr er e o g n l it on : l e as t s q川盯e s uP P o rt v e ct or m a c h i n e ( L S - SV M ) ; Z em ik e m o m en t: e l as it e m e s h