Զѧ
ACTA AUTOMATICA SINICA
1997 23 1 Vol.23 No.1 1997



ڶֽͨƥıʼо
֡ӭΪ
ժҪʼͨдַдждݵһż.ʼĹؼȡӳдıʼ.ʼʻλá򡢴ϵȣǿͨͼֽͨȡͱ.һڱʼĶֵͼַֽͨⷽıʻȽзֽ⣬ȻÿͼƵֽ.÷ֽĲźֵΪʼƥ䷽дʶ𣬵õ˺ܺõʵ.
ؼʡʼ𣬱ʼֽͨ⣬ֽ⣬Ƶֽ⣬ƥ䣬дʶ.
WRITER IDENTIFICATION BY MULTICHANNEL
DECOMPOSITION AND MATCHING
LIU CHENGLINLIU YINGJIANDAI RUWEI
(Institute of Automation, Chinese Academy of Sciences, Beijing 100080)
Abstract By writer identification (WI), the writer of a handwritten document is detected by analyzing the writing style. The crucial stage in WI is the extraction of individual features which reflect the writing style. There are many informative individual features, in which are the stroke position, direction, and the collocation. Some features can be represented simutaneously by multichannel decomposition (MCD) of images. In this paper, a MCD approach of binary images for WI is presented. The image is decomposed into directional subimages according to strokes direction, and then each subimage is further decomposed into several frequency bands. Finally, the sampled values of subband images are used as individual features for WI. Promising results have been achieved in the writer recognition experiments by feature matching.
Key words Writer identification, individual features, multichannel decomposition, directional decomposition, frequency band decomposition, feature matching, writer recognition.
1ǰ
ʼ(Writer Identfication, WI)1ͨͱȽдʼдждݵһſѧͼ.ʼķĶȡɷΪ12ݣı(text independent)ı(text dependent).ıӴַȡַ޹ص粼ַ̬֡.ı淽ӼʼͲοʼѡͬĵ(Ϊ)бȽ.ıȣı淽Զַ״дķ˿ԵõߵļʺͿɿ.һµıʼ𷽷ַͨͼжֽͨƥõıʼ.
ʼо30꣬ȡһĳɹ˺ܶ෽1.ʼıҪб任(׷)طγֱͼʶֱͼ.ı淽Ҫб׼ģη3ݡ߶θ߽ط2ݡģʽƵʷ4ݡֱָͼ4ݡʻƥ䷨5ݵ.ܵ˵Ŀǰʼˮƽ̫ߣкܶûн.ʼоĿΪʵֱʼԶобӦñǷֶεʵ.봫ͳģʽʶⲻͬǣʼǵд˺õַǲģ˲ܽѧϰͶȡͱȽб˸ߵҪ.
ʼĹǶԼʼͶ˵ĲοʼбȽϣҳджΪʼд.Ϊ˵õɿļʼһֿеķǣʼвοʼбȽϣʼƶȣȻƶȴС˽ŶӣжƶΪʼдˣƶȽϴΪصɶȻ˹רҽз϶.ȽݱʼʱҳʼͲοʼеͬ()ֵƶȣȻ϶ֵƶȵõʼ֮ƶ.ֵıȽһƥ̣ͨƥƶȻ.Ϊ˱ڽ϶ֵϢֱͬȽϽ(ƶȻ)Ӧһ(ֵַ޹).
һµֱȽϷ˼ȶԼַͲοַͼֱжͨ(Ƶ)ֽ⣬ӴͼĲֵΪʼȻ֮ľ룬ݴ˾дʶ.ַͼĶֽͨǷַֽƵֱֽеģȷٶȿص.
2ֽͨĻ˼
дַıʻλáʻʹϵǷǳҪıʼַͨͼжֽͨ(Multichannel Decomposition, MCD)Աȡͱʾ.ʻ֮ĴϵԿһԣӦһƵ׷ֲ.MCDźŷֽɲͬƵźţӶͬƵ׵źвͬķֽ.MCDƵѡзԣռƵʷַֺʻһӦϵ˲ͬʻַͼвͬķֽ.MCDڿռͿռƵͬʱзֱ˱ʻλϢҲڷֽͼзӳ.ȡַıʻбʼıף2ݵ߶θ߽طף6ݵγֱͼЩȡȫֵģʻҲûʽ.ף56ݵķֱָͼ(Directional Index Histogram, DIH)ʽ˱ʻϢûнƵֽ.ķͬʱַͼзƵֽ⣬˱ʻλϢ൱һֿռ/ռƵϱʾ7.
źŵĶֽͨͱʾкַܶõDOGʾ78ݡGabor任С任911ݵ.DOGʾһִͳźŶֽͨⷽͨͬĵͨ˲֮õͨ˲Ĳûзѡ.Gabor任С任ͬʱƵʺͷѡԣǽӵźŷ.ڶֽͨ˼źŷַʶǩ֤õ˳ɹӦã1213ݣȻGabor任С任ѧȽϳ죬ڼӾģʽʶҲȡ˳ɹӦã̫.ַͼĴǿԴٵ.ڱʼеĶֽͨǲ÷ֽDOGֽϵķ.DOG˲ǸͬԵģǿƵֽ֮ǰñʻķԶַͼзֽ⣬ȻٶÿͼƵֽ.ķֽƵͷѡ.ַַֽⷽͼص㣬Gabor任С任ԣ˼.
ͼֽͨĽǳߴ粻ģַͼķֽ֮ǰҪгߴһ.ڱʼУַͼĹһҪ󱣳״䣬ˮƽʹֱıͬ.һĵС6464.
3ַͼֽ
ֱָͼ(DIH)4һķģƥʶ𷽷ַʶͱʼжõ˳ɹӦ.DIHֻ˱ʻϢԿ˷ʻȶֵӰ.ǣֻϢõľֲʻƵ׷ֲ̫ȶҪͬʱͷзֽ.ַͼķֽǣǰ(ʻ)ֲʻһֵ(ˮƽԽǡֱҶԽǣֱ0123ʾ)Ȼĸıʻֱ䵽Ӧķͼȥ.
ķֽDIHķĹϵ跽ֵ巽.һ4-Ϊ0ǰ.ֻһ4-Ϊ0ʱ(ͼ1(a)ʾ)õıʻΪ04-㷽Ĵֱ.4-Ϊ0ʱΪ04-(ͼ1(b)ʾ)ʻߵĴֱΪ04-(ͼ1(c)ʾ)ʻߵķ.4-Ϊ0ʱ(ͼ1(d)ʾ)õıʻڲΪ04-㷽Ĵֱ.ĸ4-㶼Ϊ0ʱɿ㣬迼.


ͼ1ṹ(XΪ10)
ڷ(ڲ)ʻȷҳõı㣬Ըñڲ߷(ĸֵ)ĴֱΪʻ.Ѱڲþ任ɣ˼ǺС.õľ任ǷŷϾ任(Signed Euclidean Distance Tranform, SEDT)14ݣԸԵǰڵ꣬ӶԵõλã֮Լߵķ.ΪʡƪSEDTľ巽ﲻ.õÿڲ(LX,LY)ߵķǵ
(1)
õǦȵֵ
(2)
ڲıʻ(ߵĴֱ)
(3)
õбʻķַͼf(x,y)ֽĽ(k=0,1,2,3Ӧĸ)
(4)
4Ƶֽ⣺˲Ͳ
ֽ󣬶ÿͼеͨ˲ʹͨ˲.ͨ˲ԲԳƵGauss˲Ӧ
(5)
˲ͼźӦľ.Gauss˲ƵתƺΪ
(6)
Сu12Цx.Gauss˲Ĵuȣ߶Ȧxɷ.ͬĵͨ˲Եõͨ˲(DOG)
(7)
У߶ȡ12һ㡡2=21  2=18.ǵʵͬĵͨ˲ͼ˲ȡͨźźһͨź(Ƶ)ڱʼ.ͨźǶͨźõ.
ͼ15ݣźŵĲƵӦڻźƵʵ2.˵ˮƽʹֱƵumaxvmaxƵ
fx2umax, fy2vmax.(8)
ʹƵʵźƵʵ2.ƵźŵĴɽƵԵͨ˲Ĵ.Gauss˲޳弤Ӧ˲(IIR)ļҲǽƵģǰźŴ󲿷Ƶ.㣬ǷԦuΪƵźŵĴȽϺʣ16.˲ĲźŵĲ֮Ҳһϵ.ΪtsƵΪ fs=1/ts,fsu֮ڹϵ fs2uu=,˿Եõͨ˲Ĳ
x=ts/С(9)
ʵУͨ˲ӦĲڷֱ81632(9)ʽõǵĳ߶ȷֱǦ1 =2.546,2=5.093ͦ3=10.186,ǵӦֱʾΪ g1(x,y)g2(x,y)g3(x,y)ͨ˲ͼֱʾΪ f1(x,y) f2(x,y)f3(x,y).f1(x,y)Ĳַͼ88.f2(x,y)ĲӦ44ģҪͨ˲֮ f1(x,y)-f2(x,y)Щ f1(x,y)ͬ64ͬҪf2(x,y)ֵ. f3(x,y)ĲӦ22ģͬҲҪ88ֵ.
ַkϵͼ f(k)(x,y)k=0,1,2,3,ͨ˲ֱõͬƵĵͨźf(k)1(x,y)f(k)2(x,y)f(k)3(x,y).ÿźŽƵֽ⣺
f(k)1b(x,y)f(k)1(x,y)f(k)2(x,y)(10)
f(k)2b(x,y)f(k)2(x,y)f(k)3(x,y).(11)
f(k)1b(x,y)f(k)2b(x,y)f(k)3(x,y)88ֵf(k)2b(x,y)ƽõ44ֵf(k)3(x,y)ƽõ22ֵ.Щֵַĺڵһ(ʻȵӰ)Ϊʼ.ά4(64+16+4)=336ߡСƵֱ2566416ֵ˳ųʸf.
5ƥľ
ַʼ֮ľ룬ҪǵƵľ룬ҪͬƵľ.ƵľʱӦǲͬƵģʽɷԵĹ.ǣڲַͬźŲͬƵռܴԼҲܲһ.磬ַƵ࣬ʱҪǵƵãַĸƵ࣬ʱƵ͸ƵҲȽϴ.ˣڽϲͬƵľʱֻƵźռضԾмȨ.ַʸf1 f2(12)ʽþ뿼˲ͬƵźԼ׵ĴСǶַһ.
(12)
Ƶ͵ƵźŵĲֱֵԱ416ΪǵĲֻиƵźŲ1/41/16.ʵ֤־ǱȽȶ.
6ʵ
Ϊ˶ֱܵȽϷм飬˻ڵֱȽϵдʶʵ.ʵǴ20дıʼıѡظƵʽϸߵ֣25.ÿֵĶѡһΪοΪ.20˵ĲοȽϣСŶӣҳдƵд(ǰλѡ).ֵıʼȷ޵ģΪ˼ϵļЧǷֱ5101520ֵľдʶ.ϵľڶ־ƽֵ.ڵֺͶֽϵļ1ʾ.ΪԱȣ1˷ֱָͼ(DIH)ʵýͬʵ̺õ.
1ڵֺͶֽϵдʶ(%)

ʶʶѡѡĺѡѡ
MCD1
5
10
15
2059.40
95.84
99.84
100
10071.41
98.12
100
100
10078.74
98.88
100
100
10082.94
99.44
100
100
10086.44
99.64
100
100
100
DIH1
5
10
15
2054.87
91.96
99.16
99.72
10068.38
96.80
99.80
100
10075.79
98.36
99.92
100
10080.33
99.16
99.92
100
10083.95
99.60
99.96
100
100

DIHнܵıʼ𷽷Чõһ֣MCDõĽֱDIH˺ܶ࣬ӱ1Կ.õֽдʶʱDIHʶ54.87%ǰѡȷ83.95%MCDʶʴﵽ59.40%ǰѡȷ86.44%.ֽʱMCDĽҲõö࣬10ֽʱǰѡȷʾʹﵽ100%15ֽʱһѡ100%DIHﵽͬҪ5.
7
һµıʼ𷽷ַĻ˼ǷֱԼֺͲοֵͼжֽͨ⣬÷ֽͼĲֵΪʼ֮ľ.ַͼĶֽͨǷַֽƵֽеģС.ǶԲͬƵϣҶԲͬ־ǹһ.ڵֺͶֽϵдʶʵ·ļǷǳõģʶԸڷֱָͼ.ֽϵдʶ֮˵ַѴﵽʵˮƽ.
߼飺֡19671989人ѧѧʿѧλ19921995ֱڱҵѧпԺԶо˶ʿͲʿѧλ.Ҫоģʽʶ˹ܡźŴʶ.
ӭУ1953ڽϾ1978ϾͨŹѧԺϵѧϰ.1982ҵܲͨŲ1987걻Ƹ¼ȡΪпԺԶְʿ.пԺԶоԱʶʵΡԺƼ˾ܾ.Ҫдʶо͸߿ƼҵйѧԺȻѧһȽȶ20ƪжר.
Ϊˣ1933ҵڱѧ.ڴԶơģʽʶ˹ܡܿƼ˼άѧо.йѧԺԺʿпԺѧѧΡʵǼƻĿϯѧҡܼоѧίԱΡԺѧλίԱἰȻίѧԱˡйô󻷾ԴרεְƸ廪ѧͷѧʮУνڡְ.ڵιѧ־࣬˶ʿʿ40.ѧǮѧɭڵֱָ£ĳЩԿѧǰͬко.
ߵλйѧԺԶо100080
ο
1ݡPlamondon R, Lorrete G. Automatic signature verification and writer identificationthe state of the art. Pattern Recognition, 1989, 22(2):107131.
2ݡ߳ӲӢ˾.2ͳ߷֥ڥȥֽˤ餭֤αʶ.(ձ)ͨѧ־1984J67-D(7):776-783.
3ݡNaske R D. Writer recognition by prototype related deformation of handprinted chracters, Proc. 6th ICPR, 1982, PP. 819-822.
4ݡYoshimura I, Yoshimura M. Off-line writer identification using ordinary characters as the object, Pattern Recognition, 1991, 24(9):909-915.
5ݡImpedovo S, et al. An off-line writer identification system based on a syntactic approach, Proc. 1990 IWFHR, pp.53-61.
6ݡAzari B. Handwriting identification by means of run-length measurements, IEEE Trans. SMC, 1977,7(12):878-881.-
7ݡJacobson L D, Wechsler H.Joint spatial /spatial-frequency representation,Signal Processing, 1988, 14:37-68
8ݡCrowley J L,Stern R M. Fast computation for the difference of low-pass transform, IEEE Trans. PAMI, 1984, 6:212-221.
9ݡMallat S G. A theory for multiresolution signal decomposition:the wavelet representation,IEEE Trans. PAMI, 1989, 11(7):674-693.
10ݡMallat S G. Multifrequency channel decompositions of images and wavelet models, IEEE Trans. ASSP, 1989, 37(12):2091-2110.
11ݡJain A K, Farrokhnia F. Unsupervised texture segmentation using Gabor filters, Pattern Recognition, 1991,24(12):1167-1186.
12ݡShustorovich A. A subspace projection approach to feature extraction:the twodimensional Gabor tranform for character recognition, Neural Networks, 1994,7(8):1295-1301.
13ݡQi Y, Hunt BR. A multiresolution approach to computer verification of handwritten signatures ,IEEE Trans. Image Processing, 1995,4(6):870-874.
14ݡYe Q-Z.The signed Euclidean distance tramsform and its applications, Proc 9th. ICPR, 1988, 495-499.
15ݡJain A K. Fundementals of Digital Image Processing, Prentice-Hall, 1989.
16ݡLiu C-L, Hao H-W, Liu Y-J. Handwritten numeral recognition by spatial filtering and neural network. Proc. ICNNSP, Nanjing, 1995.
ոڡ1995-08-09


