Investigation of Normalization Techniques and Their Impact on a Recognition Rate in Handwritten Numeral Recognition

Size: px
Start display at page:

Download "Investigation of Normalization Techniques and Their Impact on a Recognition Rate in Handwritten Numeral Recognition"

Transcription

1 S C H E D A E I N F O R M A T I C A E VOLUME Investgaton of Normalzaton Technques and Ther Impact on a Recognton Rate n Handwrtten Numeral Recognton WIESŁAW CHMIELNICKI 1, KATARZYNA STĄPOR 1 Faculty of Physcs, Astronomy and Appled Computer Scence, Jagellonan Unversty, Reymonta 4, Kraków, e-mal: weslaw.chmelnck@uj.edu.pl Insttute of Computer Scence, Slesan Techncal Unversty, Akademcka 16, Glwce Abstract. Ths paper presents several normalzaton technques used n handwrtten numeral recognton and ther mpact on recognton rates. Experments wth fve dfferent feature vectors based on geometrc nvarants, Zernke moments and gradent features are conducted. The recognton rates obtaned usng combnaton of these methods wth gradent features and the SVM-rbf classfer are comparable to the best state-of-art technques. Keywords: handwrtten numeral recognton, normalzaton technques, SVM classfer, feature vectors, OCR, geometrc nvarants, Zernke moments, gradent features.

2 54 1. Introducton The recognton of handwrtten numerals by computer has been a subject of ntensve research for about last ffty years. The problem, whch s very smple for almost every human, s extremely complcated for the machne. Hundreds of scentst developed many sophstcated systems but computers are stll unable to compete wth human capabltes. The man reason that causes recognton of characters and numerals such a complex problem s the varety of wrtng styles. Each character or numeral can look dfferently dependng on a wrtng person, lght or heavy prnts, varyng levels of care, etc. A typcal OCR system usually conssts of three man processng stages: preprocessng, feature extracton/selecton and recognton usng a classfer. Preprocessng s a very mportant factor for the next two recognton stages: feature extracton and classfcaton. It conssts of a sequence of operatons appled to the mages to prepare them for feature extracton. Usually we start wth nose removal and/or smoothng [40], document skew correcton [41], normalzaton [], slant correcton [4]. Dependng on a feature extracton method addtonal preprocessng steps mght be requred such as thnnng [43] or contour analyss. Feature extracton conssts of large varety of technques whch allow us to represent an mage as a vector of values (features). These features can be based for example on geometrc moment nvarants [1], Zernke moments [7] or gradents [9, 10]. At ths stage a feature selecton algorthm can be appled to reduce the sze of the nput feature vector to avod the so-called curse of dmensonalty problem. There s also a large number of classfers. Begnnng from parametrc statstcal classfers, neural networks, SVMs (Support Vector Machnes) and fnshng on hybrd classfers. At each stage we can choose parameters whch could affect the fnal classfcaton performance. Ths paper focuses on recognton of unconstraned handwrtten numerals and especally on the normalzaton technques. It shows the mpact of mage normalzaton on the classfcaton performance. There are several normalzaton methods (descrbed n []) whch have been mplemented to use n ths research. The experments wth these methods show how dfferent normalzaton algorthms nfluence the fnal classfcaton performance. Fve dfferent feature vectors of three types (geometrc nvarants, Zernke moments and gradent features) are used n the experments. The nfluence of normalzaton on each of these vectors s shown. In ths research the SVM classfer based on the statstcal learnng theory of Vapnk [5] was used. The MNIST (modfed NIST) dgt database was used n the experments. The NIST database was collected from specally desgned forms flled by US Census Bureau employees and Hgh school students. The number of tranng

3 55 samples and test samples are and , respectvely. Ths database s wdely used n varous character recognton researches. In recent years many feature extracton methods for handwrtten numeral recognton have been proposed. A survey of these methods can be found n [50]. Addtonally, some new features such as: stroke features [0], curvature features [10], local structure features [1] or structural and concavty features [44] are used to enhance a recognton rato. There are also many approaches to the classfcaton task of handwrtten characters, lke: statstcal technques [], neural networks and Support Vector Machnes (SVMs) [3]. The parametrc or non-parametrc statstcal classfers, the lnear dscrmnant functon (LDF), the quadratc dscrmnant functon (QDF), the nearest-neghbour (1-NN) and k-nn classfers, the Parzen wndow classfer, etc are used. They can be modfed: for example a modfed quadratc dscrmnant functon was proposed by Kmura et al. [4, 6]. Neural networks nclude the multlayer perceptron (MLP), the radal bass functon (RBF) network and the polynomal classfer. Several of these classfers have been evaluated n [7]. There are three man databases: CENPARMI [16], CEDAR [8], and MNIST [9] used n handwrtten characters recognton. They have been wdely used for valdatng the recognton performance. The CENPARMI dgt database was released by the Concorda Unversty. It contans 6000 dgts dvded nto a tran set (4000 mages 400 samples per class) and a test set (000 mages 00 samples per set). The CEDAR dgt database was released by CEDAR, SUNY Buffalo. The tranng data set contans dgt mages. The test data set contans 711 dgt mages. The last database MNIST (modfed NIST) was extracted from the NIST specal databases SD3 and SD7. In ths database the szes of tranng and test sets are and mages, respectvely. Tab. 1. Results on CENPARMI database Method Recognton rate Franke [33] 97.60% Suen et al. [30] 98.85% Lu et al. [31] 98.45% Gader et al. [34] 98.30% Lu et al. [] 99.15% Below the results acheved on these databases are presented. In the followng tables the best recognton rates obtaned usng correspondng databases are shown. The recognton rate s defned as the number of correctly recognzed mages/the total number of mages of the test dataset.

4 56 The results acheved usng the CENPARMI database are presented n Tab. 1. The recognton rates vary from 97.6% to 99.15%. Some of these results have been receved usng multple classfers for example Suen et al. [30], but even better results 99.15% can be acheved usng a sngle classfer as n []. Tab.. Results on CEDAR database Method Suen et al. [30] 99.77% Lu et al. [8] 98.87% Ca et al. [35] 98.4% Oh et al. [36] 98.73% Recognton rate The lst of results obtaned usng the CEDAR database s presented n Tab.. The recognton rates are n the range from 98.4% to 99.77%. The best result was acheved by combnng multple classfers n [30], the best result usng a sngle classfer was 98.87% [8]. The thrd database, MNIST, s most wdely used for evaluaton of recognton algorthms. Some results on ths database are presented n Tab. 3. Here the recognton rates vary from 98.3% to 99.41%. The hghest accuracy was gven by Teow and Loe usng SVC on drecton and stroke-end features [8]. The same result 99.41% was acheved by Lu et al. [] on the NIST database (MNIST s a modfed NIST). Tab. 3. Results on MNIST database Method Teow et al. [8] 99.41% Mayraz et al. [37] 98.3% Dong et al. [38] 99.01% Belonge et al. [39] 99.37% Lu et al. []* 99.41% Recognton rate The rest of ths paper s organzed as follows. Secton ntroduces the normalzaton strateges, Secton 3 descrbes features extracton methods used n ths paper. Secton 4 shortly descrbes foundatons of the SVM classfer. Secton 5 presents expermental results, whle Secton 6 conclusons and future work.

5 57. Normalzaton technques Normalzaton s a process that changes dfferent mage parameters to obtan more convenent values. Normalzaton technques are used generally to reduce the wthn-class varaton. For example the ntensty normalzaton tres to equalze ntensty of the character mages, the perspectve transformaton may correct the mbalance of the character wdth [3], the moment normalzaton tres to correct the rotaton or slant [64] and the rato normalzaton changes the character aspect rato. Normalzaton s consdered to be the most mportant preprocessng factor for character recognton [55]. In most experments consderng feature extracton and classfcaton of characters a square standard plane wth the fxed N N dmensons s used. All orgnal character mages, usually of dfferent szes, are mapped onto ths plane. From ths standard plane a feature vector s extracted. The algorthms descrbed below are used to perform these mappngs. Let W1 and H1 denote the wdth and the heght of the orgnal character mage, respectvely. Then the aspect rato of the orgnal mage s defned as: mn( W1, H1) R1. (1) max( W, H ) Smlarly, let W and H denote the wdth and the heght of the normalzed mage, respectvely. Then the aspect rato the normalzed mage s defned as: 1 1 mn( W, H ) R. () max( W, H ) Both R1 and R are n the range [0, 1), as t follows from Eqs (1) and (). In the descrbed experments an aspect rato adaptve normalzaton (ARAN) strategy [] s used. In ths strategy the normalzed aspect rato R s calculated based on the orgnal aspect rato R1 usng dfferent mappng functons. So usng ARAN normalzaton a character mage s ftted nto a new normalzed plane H W, and then ths plane s shfted to overlap the standard plane. Dmensons of the normalzed plane are calculated as follows. It s assumed that one dmenson of the normalzed plane flls one dmenson of the standard plane: N = max(h, W), the other dmenson s calculated usng the aspect rato R and then centred on the standard plane (see Fg. 1). The transformaton to the standard plane can be descrbed usng a coordnate mappng. Let us denote the orgnal mage as f(x, and the normalzed mage as g(x, y ). Then the normalzed mage can be generated by coordnate mappng g(x, y ) = f(x,. We can use the forward mappng or the backward mappng. These mappngs are gven by:

6 58 Fg. 1. (a) An orgnal mage, (b) A normalzed mage on the standard plane and x x( x,, y y( x, x x( x,, y y( x, respectvely, where x, y are the coordnates of the normalzed mage and x, y are the coordnates of the orgnal mage. The mappngs used n the experments are presented n Tab. 4. When the forward mappng s used, x and y are dscrete, but x and y can be real values. Smlarly, when the backward mappng s used, then x and y are dscrete, but x and y can be real values. Moreover, n the forward mappng, the mapped coordnates x, y usually do not fll all pxels n the normalzed plane. So the coordnate dscretzaton or pxel nterpolaton s necessary. Fg.. An example of the backward mappng

7 59 The dscretzaton algorthm s qute smple. The mapped coordnates (x, y ) or (x, are approxmated by the closest nteger numbers ([x ], [y ]) or ([x], [y]). Then n case of the forward mappng, the dscrete coordnates (x, y ) scan the pxels of the orgnal mage and the pxel value f(x, s assgned to all pxels ranged from ([x (x, ], y (x, ]) to ([x (x+1, y+1)], [y (x+1, y+1)]). In case of the backward mappng dscretzaton s trval. In the descrbed experments grey-scaled mages are used, so the nterpolaton algorthm must be used. In case of the backward mappng the mapped pxel (x, s surrounded by four dscrete pxels. The grey level g(x, y ) s a weghted combnaton of the four pxel values. It s graphcally presented n Fg.. In the forward mappng every pxel n the orgnal mage and the normalzed mage are treated as squares of a unt area. The unt square of the orgnal mage s mapped to a rectangle n the normalzed plane. It s llustrated n Fg. 3. Next, each unt square overlappng the rectangle s gven a grey level value proportonal to the overlappng area. Fg. 3. An example of the forward mappng Two mappngs used n the experments are presented n Tab. 4. When usng the moment-based normalzaton method one dmenson may go beyond the standard plane. In ths case the mage part outsde the standard plane s cut off. Lnear mappng Moment mappng Tab. 4. Normalzaton methods used n experments Forward mappng Backward mappng x x x x/ y y y y/ x ( x x c ) x c x ( x x c ) / xc y ( y y c ) y c y ( y y c ) / y c where:

8 60 W / W1, H / H1 the centre of gravty of the mage s gven by mpq denotes the geometrc moments:, x m /, c 10 m00 y c m 01 / m00, m pq x y x p y q f ( x,, x c, y c denote the geometrc centre of the normalzed plane gven by: x c W /, y c H /. In the descrbed experments, all normalzaton functons are mplemented by the backward mappng. Fg. 4 shows samples of the normalzed mages, correspondng to all normalzaton functons. The sze of the standard plane s 3 3. In the descrbed experments there are several normalzaton methods mplemented as descrbed above. The used normalzaton functons are lsted n Tab. 5. Tab. 5. Lst of normalzaton functons used n the experments Symbol Descrpton Aspect rato N0 Lnear normalzaton wth fxed aspect rato R 1 N1 Lnear normalzaton wth preserved aspect rato N Lnear normalzaton wth square root rato R R 1 R R 1 N3 Lnear normalzaton wth cube root rato R 3 R 1 N4 Lnear normalzaton wth fxed aspect rato* R 9 0. N5 Lnear normalzaton wth square root of sne of R sn( / R1 N6 N7 aspect rato Moment normalzaton wth preserved aspect rato Moment normalzaton wth square root rato R R 1 R R 1 N8 Moment normalzaton wth cube root rato R 3 R 1 N9 Moment normalzaton wth fxed aspect rato* R 0. 9 N10 Moment normalzaton wth square root of sne R of aspect rato * The aspect rato obtaned usng test procedure from range [0.4, 1). ) sn( / 1) R

9 61 Fg. 4. Normalzed mages 3. Feature extracton In the descrbed experments three types of features are used: geometrc moment nvarants, Zernke moments and gradent features. The moment nvarants are known to be nvarant under rotaton, translaton, scalng and reflecton. The Zernke moments are nose reslent. The gradent features are easy to extract and gve the hgh performance and dscrmnatve power as well Geometrc moment nvarants Frst type of features whch were used are geometrc moment nvarants. These features extract global propertes of the mage such as the shape area, the centre of the mass, the moment of nerta, and so on. In these experments a feature vector smlar to presented n [46] s used, but modfed and extended to a 98D vector. Gven a grey-scale mage of the sze M N, the regular moments of order (p + q) are defned as: pq N M m x 1 j1 p y q j f ( x, y From the above translaton-nvarant central moments can be obtaned by placng the orgn n the centre of gravty. where pq N M ( x 1 j1 x) p ( y j q j ). f ( x, y j ),

10 6 Hu showed that: are scale-nvarant. pq m10 m01 x, y. m m pq 00 pq ( 1) 00 / 00, p q Fnally, rotaton-nvarant feature can be constructed. In ths paper seven nvarants were used as follows: 1 0 0, ( 0 0) 4 11, ( 3 ) ( ( 30 1) ( 1 03) ( )( )(( ) 3( ) ) ),, )( 1 03)(( 30 1) ( 1 03) ), 6 ( 0 0)(( 30 1) ( 1 03) ) 4 11( 30 1)( (3 1 03)( 03 1)(( 30 1) 3( 1 03) 1 30)( 1 03)(3( 30 1) ( 1 03) ). (3 ) (3 ), As these features are rotaton-nvarant there wll be a problem wth recognzng some numerals,.e. 6 and 9 or and 5. As a matter of fact the recognton rate usng these seven features was only about 45%. To avod ths problem the mage was dvded nto 4 and 9 square regons and all seven geometrc nvarants were extracted from every regon. Ths gves a (4 + 9) * = 98D feature vector. Ths feature vector s denoted as GMI. 3.. Zernke moments A Zernke moments concept was frst ntroduced by Teague n 1980 [49]. Compared to the geometry moment nvarants Zernke moments are computatonally expensve, but have several advantages: they are orthogonal, rotaton nvarant and nose reslent. Addtonally, they have one nterestng feature, the orgnal mage can be reconstructed from these moments. They have been used to bnary pctures because they are not nvarant due to

11 63 contrast. Ths drawback can be easly avoded usng the grey-scale normalzaton. Complex Zernke moments are constructed usng a set of complex polynomals whch form a complete orthogonal bass set defned on the unt dsc. These polynomals are defned as below: where V nm ( x, R nm ( x, e jm tan 1 ( y / x) j 1, n 0, n m s even and ( nm) / s n / s ( 1) ( x y ) ( n s)! Rnm( x,. s0 n m n m s!! s! s Then Zernke moment of order n and repetton m s defned as: n 1 f ( x, ( V ( x, ) * Anm nm, x y 1, (3) m where * denotes a complex conjugate operator, n m s even and m n. It s nterestng that the orgnal mage can be reconstructed usng the formula: N f ( x, lm A V ( x,, (4) N n0 m where the sum s taken for all m n and n m s even. The ampltudes of Zernke moments Anm are rotaton nvarant. Invarance to the scale and translaton can be obtaned by shftng and scalng the mage before the computaton of Zernke moments. The normalzaton algorthms used n these experments guarantee that all mages are shfted and scaled. There are two feature vectors based on Zernke moments used. The frst vector ZM1 conssts of all 47 frst ampltudes of Zernke moments from Z 0, 0 to Z1,1, and the second vector ZM conssts of 4 ampltudes chosen only as descrbed n [7],.e: Z0,0, Z,0, Z3,1, Z3,3, Z4,0, Z4,, Z5,1, Z5,3, Z5,5, Z6,0, Z7,1, Z7,3, Z7,5, Z8,4, Z8,6, Z9,5, Z9,7, Z10,, Z10,4, Z11,1, Z11,5, Z11,7, Z1,0. nm nm, Fast algorthm to compute Zernke moments The algorthm computng Zernke moments wth the use of Eq. (3) wll be very neffcent. As a matter of fact t s useless accordng to the database sze. So there must be found a far more effcent method. Let us notce that under polar coordnates the above formula can be expressed as:

12 64 A nm n 1 x y R nm ( r) e jm f ( r, ), (5) where r x y, tan 1 ( x / and R nm ( r) ( nm) s0 / s ( 1) ( n s)! r n m n m s!( s)!( s)! Then magnary and real components of Zernke moment can be calculated as: n C nm Rnm( r) cos( m ) f ( r, ) x y. ns (6) (7) and S nm n x y R nm ( r) sn( m ) f ( r, ). (8) The character mages are dscrete and usually relatvely small (n these experments 3 x 3) so the number of dfferent values of r s relatvely small. Let us denote the pxel n the centre of the mage as level 0, and the next 8 neghbourng pxels as level 1 and so on. On each level there wll be level+1 dfferent values of r. The number of levels wll be: n / ( n / 1). 1 Fg. 5. All possble values of r for a 7 x 7 mage

13 65 The total number of dfferent values of r wll be: 1/8(n + 6n + 8). Usually only a few dozen of frst Zernke moments are used, so all values of Rnm(r) can be computed usng Eq. (5) for all possble values of r, m, n. For example n the descrbed experments there are 3 x 3 mages and 47 frst Zernke moments are used. It gves 153*47 = 7191 dfferent values for all mages. Precomputng all these values has a great mpact on the algorthm effcency because all values of Rnm are calculated only once for all mages (the MNIST database used n the experments has = mages). Values of sn(mθ), cos(mθ) (or even Rnm(r)sn(mθ), Rnm(r)cos(mθ)) can be precomputed as well Gradent features The gradent features can be easly used to grey-scale mages and are robust aganst mage nose and edge drecton fluctuatons. Addtonally, the gradent can be computed by usng the Sobel operator, whch has two masks for the gradent components n horzontal and vertcal drectons. So t can be effcently extracted from the mage. The gradent gves us the magntude and the drecton of the greatest change n ntensty n the neghbourhood of a pxel. The Roberts [10] and Krsh [65] operator have also been used n the lterature. Fg. 6. Sobel masks used to compute gradents The Sobel operator s used to compute gradent components as follows: g x ( x, f ( x 1, y 1) f ( x 1, f ( x 1, y 1) f ( x 1, y 1) f ( x 1, f ( x 1, y 1) (9)

14 66 g y ( x, f ( x 1, y 1) f ( x, y 1) f ( x 1, y 1) f ( x 1, y 1) f ( x, y 1) f ( x 1, y 1). (10) Then, the gradent magntude s calculated as: and the gradent drecton as: A( x, g ( x, g ( x, (11) ( x, tan x 1 g ( g y x y ( x, ). ( x, The complete gradent map may contan some nose nformaton. Especally when grey-scale mages are used. To avod these spurous gradents a smple flterng algorthm s proposed an adaptve gradent thresholdng. In the frst step the average gradent magntude s computed over the whole mage and then ths value s used to flter our gradent map. Formally: g avr x y A( x, M N, where M, N are dmensons of the mage. Then and A( x, A( x, 1 ( x, ( x, 1 A( x, A A( x, A A( x, A A( x, A avr avr avr avr. The gradent drectons are real values from range [0, 360). To extract a feature vector they are quantzed nto a small number of nteger values. There are 1 nteger values used representng gradent scopes: [0, 30), [30, 60), [60, 90) and so on. Next, the gradent map s dvded nto 4 x 4 parts. Then, a percentage of pxels wth the drecton of gradent quantzed to value K = 1,,..., 1 s computed n each part. Hence the total number of features wll be = 19D. Ths feature vector s denoted as GF. The second feature vector denoted as GFC s also based on gradent features. There are 10 crossng lne features added to the prevous vector GF. The crossng lne features are extracted n the followng steps. Frst, the centre of gravty of the mage s found, then the horzontal and vertcal lne are drawn through ths pont, and fnally two extra lnes on each sde of the horzontal and vertcal lne are added wth equal margns. The crossng lne

15 67 feature s the number of ntersecton ponts wth the mage. For example for the numeral shown n Fg. 7 there are two vectors: (1,1,1,1,1) and (1,,,,1). Fg. 7. Crossng lne features (a) horzontal, (b) vertcal 4. The SVM classfer The Support Vector Machne (SVM) has been proposed by Vapnk n [5]. The SVM technque has been used n dfferent applcaton domans and has outperformed the tradtonal technques n terms of generalzaton capablty. Contrary to the tradtonal technques whch try to mnmse the emprcal rsk (the classfcaton error on the tranng data) SVM mnmses the structural rsk (the classfcaton error on data never seen before). The classfcaton task s to predct whether a test sample belongs to one of two classes. In a feature space ths corresponds to fndng a hyperplane whch separates these two classes. There s an nfnte number of such hyperplanes, so among the possble choces, the SVM classfer selects the one for whch the dstance of the hyperplane from the closest feature vectors (the margn ) s as large as possble. Ths hyperplane s called an optmal separatng hyperplane. Let us consder a classfer whose decson functon s gven by: T f ( x) sgn( x w b), where x denotes a feature vector and w s a weght vector. The problem s separable when there exst w and threshold b such that: T y ( x w b) 1, 1,,, m. To maxmze the margn we must mnmze 1/ w Ths problem leads to a so-called dual optmzaton problem, whch s such that L D N 1/ N 1, j. ( x j T x j ),

16 68 0, 1,,, N and 0. Ths leads to a hyperplane decson functon: f ( x) sgn( y ( x support vectors T N 1 y x) b), (1) where x are support vectors wth Lagrangan non-zero multplers α. The support vectors are the feature vectors whch le on the margns. Ths s an advantage of ths approach because only a small number of vectors s used to compute a resultng classfer. In a real lfe problem t s unlkely that a hyperplane wll exactly separate the data. To deal wth ths problem the soft margn hyperplanes are used. A set of varables ξ representng errors (.e. the vectors whch le nsde the margn) and a parameter C whch determnes a trade-off between margn maxmzaton and error mnmzaton are ntroduced. In Eq. (1) a dot product of the nput vectors s used. So we can apply some trck to calculate the dot product of the vectors n the feature space usng a kernel functon. It allows us to create a decson functon that s nonlnear n the nput space, but s lnear n the feature space,.e.: f ( x) sgn( y K( x, x) b), support vectors (13) where K(x, x) s a kernel functon. Typcal kernel functons are: 1. Lnear kernel: T K x, x) x x, (. Polynomal kernel: T d K( x, x) ( x x c), 3. RBF (Radal bass kernel) Kernel: K x, x) x x, 0, ( 4. Gausan RBF Kernel: K( x, x) exp( x ) /, x T 5. Sgmod kernel: K( x, x) tanh( x x c). The last problem whch must be solved s that handwrtten numeral recognton s a mult-class problem and the SVM s a bnary classfer. There are two commonly used solutons. The frst s WTA (wnner takes all) strategy. In ths approach we buld N classfers for N classfcaton problems: one class versus all other classes. Another approach s to buld n(n 1)/ classfers for each par of classes, then use MVS (majorty votng scheme) strategy.

17 69 5. Expermental results In ths secton recognton accuraces obtaned n the experments are presented. The results are presented n Tab. 6. In the rows there are fve dfferent feature vectors (descrbed n Sect. 3) and n the columns N0 N10 are normalzaton methods used (see Tab. ). In the frst column WN there s accuracy obtaned wthout any normalzaton. The handwrtten dgt database MNIST descrbed n Secton 1 s used n these experments. Some mages of the tranng dataset are shown n Fg. 8. Ths database s dvded nto two datasets: the tranng dataset ncludng samples and the test dataset ncludng samples. As a SVM-rbf classfer s used n the experments there s an extra dataset necessary for valdaton purposes (to fnd C and γ parameters). So the tranng dataset was dvded nto two sets: samples for the tranng set and samples for the valdaton set. For these three datasets feature vectors are generated usng normalzaton methods lsted n Tab. 5. For normalzaton methods N4 and N9 a specal procedure was used to fnd the best rato R. The 1 feature vectors are generated usng aspect ratos from 0.4 to 0.95 wth step The best result (for aspect rato 0.9) s presented n Tab. 6. The extra feature vector set s generated for orgnal mages. Fg. 8. Examples from MNIST database

18 70 Before usng the classfer all feature vectors are lnearly scaled nto [ 1, 1] range. The man advantage of usng ths scalng s to avod attrbutes n greater numerc ranges domnatng those n smaller numerc ranges. Another advantage s to avod numercal dffcultes durng the calculaton. Because kernel values usually depend on the nner products of feature vectors, e.g. the lnear kernel and the polynomal kernel, large attrbute values mght cause numercal problems. In Tab. 6 the results of the experments are shown. There are feature vectors n rows and normalzaton methods n columns. Tab. 6. Recognton rates obtaned on dfferent feature vectors WN N0 N1 N N3 N4 N5 N6 N7 N8 N9 N10 GMI ZM ZM GF GFC , Geometrc moments nvarants are the poor feature vector. But as can be seen n [46] and n our experments the results are close to the best features vectors based on gradent and drectonal features. Possbly, f we add extra features to these vectors,.e. concave features or crossed lnes features, the result would be even better. Zernke moments are better than geometrc moments, but are not so good as gradent features. In ths paper we manly focus on normalzaton methods and as a matter of fact we do not nvestgate the feature vectors. Maybe methods presented n [7] can be extended and yeld to better results. The best results are acheved usng gradent features. The extended GFC vector leads to even better results. The geometrc moment nvarants and Zernke moments appear slghtly worse, but the results are promsng. Perhaps combnng these feature vectors wth other features wll lead to better recognton ratos. GF and GFC vectors are most promsng. Consderng that they are easy to extract and easy to understand they seem to be good choce for future work. To examne how normalzaton methods nfluence recognton rates all feature vectors are extracted from the dataset also wthout normalzaton (the column WN n Tab. 6 shows the recognton rate on ths vector). Tab. 7 presents relatve recognton rates defned as follows: RRR j 100 Rj R *100,

19 71 where Rj s recognton rate obtaned on -th feature vector usng j-th normalzaton and R s recognton rate obtaned on -th feature vector wthout any normalzaton. Ths measure shows how a normalzaton method contrbutes to achevng the optmal recognton rate. For example RRR = 100 means that ths normalzaton method leads to the maxmal recognton rate = 100%, RRR = 0 means that the correspondng normalzaton method brngs no advantage to the result and values less than zero mean that the normalzaton deterorates the recognton rato. Tab. 7. Relatve recognton rates N0 N1 N N3 N4 N5 N6 N7 N8 N9 N10 MIs ZM ZM 43, GF GFC Conclusons and future work In ths paper several normalzaton methods and fve feature vectors are compared on the MNIST database. The recognton results show that the moment normalzaton functons N7 yeld the hghest recognton rates. The results obtaned usng moment normalzaton functons N10 and N9 are also very good. Generally, the normalzaton s nfluental to the recognton performance for both dmenson-based and moment-based normalzaton. It s nterestng that preservng the aspect rato or forcng the aspect rato to one leads to substantally worse results. There are fve sets of feature vectors tested. The best results are acheved usng gradent features, but t can be seen that other feature vectors are not qute useless. The results are promsng. Maybe some extenson of these feature vectors could lead to results comparable wth gradent features. The reported results provde useful nsghts for selectng a sutable normalzaton algorthm n developng recognton systems. There are also nterestng results n experments usng dfferent aspect ratos. Testng one class versus all others shows that there s no unversal aspect rato optmal for all classes. Dfferent aspect ratos are optmal for dfferent numerals. For example for numeral 1 the best aspect rato s 0.4 and

20 7 for numeral 0 the best aspect rato s Ths observaton s useless n ths experment, because there must be one classfer for all numerals, but n the future work t can be used for buldng a multple classfer soluton. The experments descrbed n ths paper are focused on normalzaton. The results show that ths preprocessng technque has a great mpact on the fnal recognton rate regardless of the feature vector used. The next step s to fnd even better feature vectors whch n conjuncton wth these normalzaton technques wll lead to even better recognton rates. 7. References [1] Xu D., L H.; Geometrc moment nvarants, Pattern Recognton 41, 008, pp [] Lu Ch.L., Nakashma K., Sako H., Fujsava H.; Handwrtten dgt recognton: nvestgaton of normalzaton and feature extracton technques, Pattern Recognton 37, 004, pp [3] Chang C.-C., Ln C.-J.; LIBSVM: a lbrary for support vector machnes, software avalable at: [4] Lauer F., Suen Ch.Y., Bloch G.; A tranable feature extractor for handwrtten dgt recognton, Pattern Recognton 40, 007, pp [5] Zhang W., Tang Y.Y., Xue Y.; Handwrtten Character Recognton Usng Combned Gradent and Wavelet Feature, Internatonal Conference on Computatonal Intellgence and Securty, Vol. 1, 006, pp [6] Stąpor K.; Automatc object classfcaton, Publshng House EXIT, 005. [7] Tong X.J., Zeng S., Zhou K., Jang Q.; Hand-wrtten numeral recognton based on Zernke moment, Proceedngs of the 008 ICWAPR, pp [8] Teow L.-N., Loe K.-F.; Robust vson-based features and classfcaton schemes for offlne handwrtten dgt recognton, Pattern Recognton 35(11), 00, pp [9] Lu H., Dng X.; Handwrtten Character Recognton Usng Gradent Feature and Quadratc Classfer wth Multple Dscrmnaton Schemes, Proceedngs of the Eghth ICDAR, 005, pp

21 73 [10] Sh M., Fujsava Y., Wakabayash T., Kmura F.; Handwrtten Numeral Recognton usng gradent and curvature of gray scale mage, Pattern Recognton 35(10), 00, pp [11] Crstann N., Scholkopf B.; Support vector machnes and Kernel methods: the new generaton of learnng machnes, AI Magazne 13(3), 00, pp [1] Lu Ch.L., Nakashma K., Sako H., Fujsava H.; Handwrtten dgt recognton: benchmarkng of state-of-the-art technques, Pattern Recognton 36, 003, pp [13] Sh M., Fujsawa Y., Wakabayash T., Kmura F.; Handwrtten numeral recognton usng gradent and curvature of gray scale mage, Pattern Recognton 35, 00, pp [14] Scholkopf B., Smola A.J.; Learnng wth Kernels. Support Vector Machnes, Regularzaton, Optmzaton, and Beyond, The MIT Press, 001. [15] Cheret M., Kharma N., Lu Ch.-L., Suen Ch.-Y.; Character Recognton Systems: A gude for students and practoners, Wley-Interscence, 007. [16] Suen Ch.Y., Nadal Ch., Legault R., Ma T.A., Lam L.; Computer Recognton of unconstraned handwrtten numeral, Proceedngs of the IEEE, 80, 199, pp [17] Srkantan G., Lam S.W., SrHar S.N.; Gradent-based contour encodng for character recognton, Pattern Recognton 9, 1996, pp [18] Arca N., Yarmna-Vural F.T.; Optcal Character Recognton for Cursve Handwrtng, IEEE Transactons on Pattern Analyss and Machne Intellgence 3, 00, pp [19] Lu C.-L., Nakashma K., Sako H., Fujsawa H.; Aspect Rato adaptve normalzaton for handwrtten character recognton, n: Advances n Multmodal Interfaces ICMI 000, T. Tan Y. Sh, W. Gao (eds.), Lecture Notes n Computer Scence 1948, 000, pp [0] Kmura et al.; Evaluaton an synthess of feature vectors for handwrtten numeral recognton, IEICE Trans. Inform. Systems E79-D(5), 1996, pp [1] Heutte L., Paquet T., Moreau J.V., Lecourter Y., Olver C.; A structural/ statstcal feature based vector for handwrtten character recognton, Pattern Recognton Letters 19(7), 1998, pp [] Jan A.K., Dun R.P.W., Mao J.; Statstcal Pattern Recognton: a revew, IEEE Transactons on Pattern Analyss and Machne Intellgence (1), 000, pp

22 74 [3] Burges C.J.C.; A tutoral on support vector machnes for pattern recognton, Knowledge Dscovery Data Mnng (), 1998, pp [4] Kmura F., Takashna K., Tsuruoka S., Myake Y.; Modfed quadratc dscrmnant functons and the applcaton to Chnese character recognton, IEEE Trans. Pattern Anal. Mach. Intell. 9(1), 1987, pp [5] Vapnk V.; The Nature of Statstcal Learnng Theory, Sprnger, New York [6] Kmura F., Shrdhar M.; Handwrtten numeral recognton based on multple algorthms, Pattern Recognton 4(10), 1991, pp [7] Lu C.-L., Sako H., Fujsawa H.; Performance evaluaton of pattern classfers for handwrtten character recognton, Internatonal Journal on Document Analyss Recognton 4(3), 00, pp [8] Lee D.-S., Srhar S.-N.; Handprnted dgt recognton: a comparson of algorthms, Proceedngs of the Thrd Internatonal Workshop on Fronters of Handwrtng Recognton, Buffalo, New York, 1993, pp [9] LeCun Y. et al.; Comparson of learnng algorthms for handwrtten dgt recognton, n: Proceedngs of the Internatonal Conference on Artfcal Neural Networks, F. Fogelman-Soule, P. Gallnar (eds.), Nanterre, France 1995, pp [30] Suen C.-Y., Lu K., Strathy N.W.; Sortng and recognzng cheques and fnancal documents, n: Document Analyss Systems: Theory and Practce, S.-W. Lee, Y. Nakano (eds.), Sprnger, Berln 1999, pp [31] Lu C.-L., Nakagawa M.; Handwrtten numeral recognton usng neural networks: mprovng the accuracy by dscrmnatve tranng, Proceedngs of the Ffth Internatonal Conference on Document Analyss and Recognton, 1999, pp [3] Naggy G., Tuong N.; Normalzaton technques for handprnted numerals, Communcatons of the ACM 13(8), 1970, pp [33] Franke J.; Isolated handprnted dgt recognton, n: Handbook of Character Recognton and Document Image Analyss, H. Bunke, P.S.P. Wang (eds.), World Scentfc, Sngapore 1997, pp [34] Gader P.D., Khabou M.A.; Automatc feature generaton for handwrtten dgt recognton, IEEE Transactons on Pattern Analyss and Machne Intellgence 18(1), 1996, pp

23 75 [35] Ca J.-H., Lu Z.-Q.; Integraton of structural and statstcal nformaton for unconstraned handwrtten numeral recognton, IEEE Transactons on Pattern Analyss and Machne Intellgence 1(3), 1999, pp [36] Oh I.-S., Lee J.-S., Suen C.Y.; Analyss of class separaton and combnaton of class-dependent features for handwrtng recognton, IEEE Transactons on Pattern Analyss and Machne Intellgence 1(10), 1999, pp [37] Mayraz G., Hnton G.E.; Recognzng handwrtten dgts usng herarchcal products of experts, IEEE Transactons on Pattern Analyss and Machne Intellgence 4(), 00, pp [38] Dong J.X., Krzyzak A., Suen C.Y.; A mult-net learnng framework for pattern recognton, Proceedngs of the Sxth Internatonal Conference on Document Analyss and Recognton, Seattle 001, pp [39] Belonge S., Malk J., Puzcha J.; Shape matchng and object recognton usng shape contexts, IEEE Transactons on Pattern Analyss and Machne Intellgence 4(4), 00, pp [40] Gonzalez R.C., Woods R.E.; Dgtal Image Processng, nd edton, Addson Wesley, 001. [41] Hull J.J.; Document mage skew detecton: Survey and annotated bblography, n: Document Analyss Systems II, J.J. Hull and S.L. Taylor (eds.), World Scentfc, Sngapore, 1998, pp [4] Yamaguch T., Nakano Y., Maruyama M., Myao H., Hanno T.; Dgt classfcaton on sggnboard for telephone number recognton, Proceedngs of the 7 th Internatonal Conference for Document Analyss and Recognton, Ednburgh, Scotland 003, pp [43] Zhang T.Y., Suen C.Y.; A fast parallel algorthm for thnnng dgtal patterns, Communcaton of the ACM 7(3), 1984, pp [44] Favata J.T., Srkantan G., Srhar S.N.; Handprnted character/dgt recognton usng a multple feature/resoluton phlosophy, Proceedngs of the Fourth Internatonal Workshop on Fronters of Handwrtng Recognton, Tape 1994, pp [45] de Olvera Jr. J.J., Veloso L.R., de Carvalho J.M.; Interpolaton/decmaton scheme appled to sze normalzaton of characters mages, Proceedngs of the 15th Internatonal Conference Pattern Recognton, Vol., Barcelona, Span 000, pp

24 76 [46] Ramteke R.J., Mehrotra S.C.; Feature Extracton Based on Moment Invarants for Handwrtng, IEEE Conference on Recognton Cybernetcs and Intellgent Systems, Issue 7 9, 006, pp [47] Cheng D., Yan H.; Recognton of handwrtten dgts based on contour nformaton, Pattern Recognton 31(3), 1998, pp [48] Tong X.J., Zeng S., Zhou K., Zhao K., Jang Q.; Hand-wrtten numeral recognton based on Zernke moment, Proceedngs of the 008 Internatonal Conference on Wavelet Analyss and Pattern Recognton, Vol. 1, 008, pp [49] Teague M.R.; Image analyss va the general theory of moments, Journal of the Optcal Socety of Amerca 70(8), 1980, pp [50] Trer O.D., Jan A.K., Taxt T.; Feature extracton. Methods for character recognton a survey, Pattern Recognton 9, 1996, pp [51] Kawamura A. et al.; On-lne recognton of freely handwrtten Japanese characters usng drectonal feature denstes, Proceedngs of the 11th Internatonal Conference on Pattern Recognton, Vol., The Hague 199, pp [5] Mor S., Suen C.Y., Yamamoto K.; Hstorcal revew of OCR research and development, Proceedngs of IEEE 80(7), 199, pp [53] Khotanzad A., Hong Y.H.; Invarant mage recognton by Zernke moments, IEEE Transactons on Pattern Analyss and Machne Intellgence 1(5), 1990, pp [54] Burges C.J.C.; A tutoral on support vector machnes for pattern recognton, Knowledge Dscovery Data Mnng (), 1998, pp [55] Gudessen A.; Quanttatve Analyss of preprocessng technques for the recognton of handprnted characters, Pattern Recognton 8, 1976, pp [56] Crstann N., Shawe-Taylor J.; An Introducton to Support Vector Machnes and Other Kernel-Based Learnng Methods, Cambrdge Unversty Press, 000. [57] Labusch K., Barth E., Martnetz T.; Smple Method for Hgh-Performance Dgt Recognton Based on Sparse Codng, IEEE Transacton on Neural Networks 19(11), 008, pp [58] Fan R.E., Chen P.H., Ln C.J.; Workng Set Selecton Usng Second Order Informaton for Tranng Support Vector Machnes, Journal of Machne Learnng Research 6, 005, pp

25 77 [59] Keerth S.S., Ln C.J.; Asymptotc behavors of support vector machnes wth Gaussan kernel, Neural Computaton 15(7), 003, pp [60] Kernel machnes web ste, [61] The MNIST database of handwrtten dgts, [6] Hsu C.W., Chang C.C., Ln C.J.; A practcal gude to support vector classfcaton, [63] Mukundan R., RamaKrshnan K.R.; Fast Computaton of Legendre and Zernke Moments, Pattern Recognton 8(9), 1995, pp [64] Casey R.G.; Moment normalzaton of handprnted character, IBM Journal of Research and Development 14, 1970, pp [65] Lee S.-W.; Multlayer cluster neural network for totally unconstraned handwrtten numeral recognton, Neural Networks 8(5), 1995, pp [66] Abuhaba I.S.I., Holt M.J.J., Datta S.; Recognton of off-lne handwrtng, Computer Vson and Image Understandng 71, 1998, pp Receved May 16, 010

Forecasting the Direction and Strength of Stock Market Movement

Forecasting the Direction and Strength of Stock Market Movement Forecastng the Drecton and Strength of Stock Market Movement Jngwe Chen Mng Chen Nan Ye cjngwe@stanford.edu mchen5@stanford.edu nanye@stanford.edu Abstract - Stock market s one of the most complcated systems

More information

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification Lecture 4: More classfers and classes C4B Machne Learnng Hlary 20 A. Zsserman Logstc regresson Loss functons revsted Adaboost Loss functons revsted Optmzaton Multple class classfcaton Logstc Regresson

More information

Face Verification Problem. Face Recognition Problem. Application: Access Control. Biometric Authentication. Face Verification (1:1 matching)

Face Verification Problem. Face Recognition Problem. Application: Access Control. Biometric Authentication. Face Verification (1:1 matching) Face Recognton Problem Face Verfcaton Problem Face Verfcaton (1:1 matchng) Querymage face query Face Recognton (1:N matchng) database Applcaton: Access Control www.vsage.com www.vsoncs.com Bometrc Authentcaton

More information

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis The Development of Web Log Mnng Based on Improve-K-Means Clusterng Analyss TngZhong Wang * College of Informaton Technology, Luoyang Normal Unversty, Luoyang, 471022, Chna wangtngzhong2@sna.cn Abstract.

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Max Wellng Department of Computer Scence Unversty of Toronto 10 Kng s College Road Toronto, M5S 3G5 Canada wellng@cs.toronto.edu Abstract Ths s a note to explan support vector machnes.

More information

Single and multiple stage classifiers implementing logistic discrimination

Single and multiple stage classifiers implementing logistic discrimination Sngle and multple stage classfers mplementng logstc dscrmnaton Hélo Radke Bttencourt 1 Dens Alter de Olvera Moraes 2 Vctor Haertel 2 1 Pontfíca Unversdade Católca do Ro Grande do Sul - PUCRS Av. Ipranga,

More information

L10: Linear discriminants analysis

L10: Linear discriminants analysis L0: Lnear dscrmnants analyss Lnear dscrmnant analyss, two classes Lnear dscrmnant analyss, C classes LDA vs. PCA Lmtatons of LDA Varants of LDA Other dmensonalty reducton methods CSCE 666 Pattern Analyss

More information

ECE544NA Final Project: Robust Machine Learning Hardware via Classifier Ensemble

ECE544NA Final Project: Robust Machine Learning Hardware via Classifier Ensemble 1 ECE544NA Fnal Project: Robust Machne Learnng Hardware va Classfer Ensemble Sa Zhang, szhang12@llnos.edu Dept. of Electr. & Comput. Eng., Unv. of Illnos at Urbana-Champagn, Urbana, IL, USA Abstract In

More information

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College Feature selecton for ntruson detecton Slobodan Petrovć NISlab, Gjøvk Unversty College Contents The feature selecton problem Intruson detecton Traffc features relevant for IDS The CFS measure The mrmr measure

More information

Bayesian Network Based Causal Relationship Identification and Funding Success Prediction in P2P Lending

Bayesian Network Based Causal Relationship Identification and Funding Success Prediction in P2P Lending Proceedngs of 2012 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 25 (2012) (2012) IACSIT Press, Sngapore Bayesan Network Based Causal Relatonshp Identfcaton and Fundng Success

More information

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network 700 Proceedngs of the 8th Internatonal Conference on Innovaton & Management Forecastng the Demand of Emergency Supples: Based on the CBR Theory and BP Neural Network Fu Deqang, Lu Yun, L Changbng School

More information

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module LOSSLESS IMAGE COMPRESSION SYSTEMS Lesson 3 Lossless Compresson: Huffman Codng Instructonal Objectves At the end of ths lesson, the students should be able to:. Defne and measure source entropy..

More information

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement An Enhanced Super-Resoluton System wth Improved Image Regstraton, Automatc Image Selecton, and Image Enhancement Yu-Chuan Kuo ( ), Chen-Yu Chen ( ), and Chou-Shann Fuh ( ) Department of Computer Scence

More information

A Set of Features Extraction Methods for the Recognition of the Isolated Handwritten Digits

A Set of Features Extraction Methods for the Recognition of the Isolated Handwritten Digits Internatonal Journal of Computer and Communcaton Engneerng, Vol. 3, No. 5, September 214 A Set of Features Extracton Methods for the Recognton of the Isolated Handwrtten Dgts S. Ouchtat, M. Redjm, and

More information

Gender Classification for Real-Time Audience Analysis System

Gender Classification for Real-Time Audience Analysis System Gender Classfcaton for Real-Tme Audence Analyss System Vladmr Khryashchev, Lev Shmaglt, Andrey Shemyakov, Anton Lebedev Yaroslavl State Unversty Yaroslavl, Russa vhr@yandex.ru, shmaglt_lev@yahoo.com, andrey.shemakov@gmal.com,

More information

Lecture 2: Single Layer Perceptrons Kevin Swingler

Lecture 2: Single Layer Perceptrons Kevin Swingler Lecture 2: Sngle Layer Perceptrons Kevn Sngler kms@cs.str.ac.uk Recap: McCulloch-Ptts Neuron Ths vastly smplfed model of real neurons s also knon as a Threshold Logc Unt: W 2 A Y 3 n W n. A set of synapses

More information

An interactive system for structure-based ASCII art creation

An interactive system for structure-based ASCII art creation An nteractve system for structure-based ASCII art creaton Katsunor Myake Henry Johan Tomoyuk Nshta The Unversty of Tokyo Nanyang Technologcal Unversty Abstract Non-Photorealstc Renderng (NPR), whose am

More information

What is Candidate Sampling

What is Candidate Sampling What s Canddate Samplng Say we have a multclass or mult label problem where each tranng example ( x, T ) conssts of a context x a small (mult)set of target classes T out of a large unverse L of possble

More information

On-Line Fault Detection in Wind Turbine Transmission System using Adaptive Filter and Robust Statistical Features

On-Line Fault Detection in Wind Turbine Transmission System using Adaptive Filter and Robust Statistical Features On-Lne Fault Detecton n Wnd Turbne Transmsson System usng Adaptve Flter and Robust Statstcal Features Ruoyu L Remote Dagnostcs Center SKF USA Inc. 3443 N. Sam Houston Pkwy., Houston TX 77086 Emal: ruoyu.l@skf.com

More information

Biometric Signature Processing & Recognition Using Radial Basis Function Network

Biometric Signature Processing & Recognition Using Radial Basis Function Network Bometrc Sgnature Processng & Recognton Usng Radal Bass Functon Network Ankt Chadha, Neha Satam, and Vbha Wal Abstract- Automatc recognton of sgnature s a challengng problem whch has receved much attenton

More information

Performance Analysis and Coding Strategy of ECOC SVMs

Performance Analysis and Coding Strategy of ECOC SVMs Internatonal Journal of Grd and Dstrbuted Computng Vol.7, No. (04), pp.67-76 http://dx.do.org/0.457/jgdc.04.7..07 Performance Analyss and Codng Strategy of ECOC SVMs Zhgang Yan, and Yuanxuan Yang, School

More information

Forensic Handwritten Document Retrieval System

Forensic Handwritten Document Retrieval System Forensc Handwrtten Document Retreval System Sargur N SRIHARI and Zhxn SHI + Center of Excellence for Document Analyss and Recognton (CEDAR), Unversty at Buffalo, State Unversty of New York, Buffalo, USA

More information

CS 2750 Machine Learning. Lecture 3. Density estimation. CS 2750 Machine Learning. Announcements

CS 2750 Machine Learning. Lecture 3. Density estimation. CS 2750 Machine Learning. Announcements Lecture 3 Densty estmaton Mlos Hauskrecht mlos@cs.ptt.edu 5329 Sennott Square Next lecture: Matlab tutoral Announcements Rules for attendng the class: Regstered for credt Regstered for audt (only f there

More information

Support Vector Machine Model for Currency Crisis Discrimination. Arindam Chaudhuri 1. Abstract

Support Vector Machine Model for Currency Crisis Discrimination. Arindam Chaudhuri 1. Abstract Support Vector Machne Model for Currency Crss Dscrmnaton Arndam Chaudhur Abstract Support Vector Machne (SVM) s powerful classfcaton technque based on the dea of structural rsk mnmzaton. Use of kernel

More information

Fault tolerance in cloud technologies presented as a service

Fault tolerance in cloud technologies presented as a service Internatonal Scentfc Conference Computer Scence 2015 Pavel Dzhunev, PhD student Fault tolerance n cloud technologes presented as a servce INTRODUCTION Improvements n technques for vrtualzaton and performance

More information

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by 6 CHAPTER 8 COMPLEX VECTOR SPACES 5. Fnd the kernel of the lnear transformaton gven n Exercse 5. In Exercses 55 and 56, fnd the mage of v, for the ndcated composton, where and are gven by the followng

More information

"Research Note" APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES *

Research Note APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES * Iranan Journal of Scence & Technology, Transacton B, Engneerng, ol. 30, No. B6, 789-794 rnted n The Islamc Republc of Iran, 006 Shraz Unversty "Research Note" ALICATION OF CHARGE SIMULATION METHOD TO ELECTRIC

More information

Mining Feature Importance: Applying Evolutionary Algorithms within a Web-based Educational System

Mining Feature Importance: Applying Evolutionary Algorithms within a Web-based Educational System Mnng Feature Importance: Applyng Evolutonary Algorthms wthn a Web-based Educatonal System Behrouz MINAEI-BIDGOLI 1, and Gerd KORTEMEYER 2, and Wllam F. PUNCH 1 1 Genetc Algorthms Research and Applcatons

More information

Support vector domain description

Support vector domain description Pattern Recognton Letters 20 (1999) 1191±1199 www.elsever.nl/locate/patrec Support vector doman descrpton Davd M.J. Tax *,1, Robert P.W. Dun Pattern Recognton Group, Faculty of Appled Scence, Delft Unversty

More information

v a 1 b 1 i, a 2 b 2 i,..., a n b n i.

v a 1 b 1 i, a 2 b 2 i,..., a n b n i. SECTION 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS 455 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS All the vector spaces we have studed thus far n the text are real vector spaces snce the scalars are

More information

Can Auto Liability Insurance Purchases Signal Risk Attitude?

Can Auto Liability Insurance Purchases Signal Risk Attitude? Internatonal Journal of Busness and Economcs, 2011, Vol. 10, No. 2, 159-164 Can Auto Lablty Insurance Purchases Sgnal Rsk Atttude? Chu-Shu L Department of Internatonal Busness, Asa Unversty, Tawan Sheng-Chang

More information

NEURO-FUZZY INFERENCE SYSTEM FOR E-COMMERCE WEBSITE EVALUATION

NEURO-FUZZY INFERENCE SYSTEM FOR E-COMMERCE WEBSITE EVALUATION NEURO-FUZZY INFERENE SYSTEM FOR E-OMMERE WEBSITE EVALUATION Huan Lu, School of Software, Harbn Unversty of Scence and Technology, Harbn, hna Faculty of Appled Mathematcs and omputer Scence, Belarusan State

More information

A machine vision approach for detecting and inspecting circular parts

A machine vision approach for detecting and inspecting circular parts A machne vson approach for detectng and nspectng crcular parts Du-Mng Tsa Machne Vson Lab. Department of Industral Engneerng and Management Yuan-Ze Unversty, Chung-L, Tawan, R.O.C. E-mal: edmtsa@saturn.yzu.edu.tw

More information

Conversion between the vector and raster data structures using Fuzzy Geographical Entities

Conversion between the vector and raster data structures using Fuzzy Geographical Entities Converson between the vector and raster data structures usng Fuzzy Geographcal Enttes Cdála Fonte Department of Mathematcs Faculty of Scences and Technology Unversty of Combra, Apartado 38, 3 454 Combra,

More information

An Alternative Way to Measure Private Equity Performance

An Alternative Way to Measure Private Equity Performance An Alternatve Way to Measure Prvate Equty Performance Peter Todd Parlux Investment Technology LLC Summary Internal Rate of Return (IRR) s probably the most common way to measure the performance of prvate

More information

MACHINE VISION SYSTEM FOR SPECULAR SURFACE INSPECTION: USE OF SIMULATION PROCESS AS A TOOL FOR DESIGN AND OPTIMIZATION

MACHINE VISION SYSTEM FOR SPECULAR SURFACE INSPECTION: USE OF SIMULATION PROCESS AS A TOOL FOR DESIGN AND OPTIMIZATION MACHINE VISION SYSTEM FOR SPECULAR SURFACE INSPECTION: USE OF SIMULATION PROCESS AS A TOOL FOR DESIGN AND OPTIMIZATION R. SEULIN, F. MERIENNE and P. GORRIA Laboratore Le2, CNRS FRE2309, EA 242, Unversté

More information

A Secure Password-Authenticated Key Agreement Using Smart Cards

A Secure Password-Authenticated Key Agreement Using Smart Cards A Secure Password-Authentcated Key Agreement Usng Smart Cards Ka Chan 1, Wen-Chung Kuo 2 and Jn-Chou Cheng 3 1 Department of Computer and Informaton Scence, R.O.C. Mltary Academy, Kaohsung 83059, Tawan,

More information

A Multi-Camera System on PC-Cluster for Real-time 3-D Tracking

A Multi-Camera System on PC-Cluster for Real-time 3-D Tracking The 23 rd Conference of the Mechancal Engneerng Network of Thaland November 4 7, 2009, Chang Ma A Mult-Camera System on PC-Cluster for Real-tme 3-D Trackng Vboon Sangveraphunsr*, Krtsana Uttamang, and

More information

How To Understand The Results Of The German Meris Cloud And Water Vapour Product

How To Understand The Results Of The German Meris Cloud And Water Vapour Product Ttel: Project: Doc. No.: MERIS level 3 cloud and water vapour products MAPP MAPP-ATBD-ClWVL3 Issue: 1 Revson: 0 Date: 9.12.1998 Functon Name Organsaton Sgnature Date Author: Bennartz FUB Preusker FUB Schüller

More information

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters Frequency Selectve IQ Phase and IQ Ampltude Imbalance Adjustments for OFDM Drect Converson ransmtters Edmund Coersmeer, Ernst Zelnsk Noka, Meesmannstrasse 103, 44807 Bochum, Germany edmund.coersmeer@noka.com,

More information

SVM Tutorial: Classification, Regression, and Ranking

SVM Tutorial: Classification, Regression, and Ranking SVM Tutoral: Classfcaton, Regresson, and Rankng Hwanjo Yu and Sungchul Km 1 Introducton Support Vector Machnes(SVMs) have been extensvely researched n the data mnng and machne learnng communtes for the

More information

Statistical Approach for Offline Handwritten Signature Verification

Statistical Approach for Offline Handwritten Signature Verification Journal of Computer Scence 4 (3): 181-185, 2008 ISSN 1549-3636 2008 Scence Publcatons Statstcal Approach for Offlne Handwrtten Sgnature Verfcaton 2 Debnath Bhattacharyya, 1 Samr Kumar Bandyopadhyay, 2

More information

A study on the ability of Support Vector Regression and Neural Networks to Forecast Basic Time Series Patterns

A study on the ability of Support Vector Regression and Neural Networks to Forecast Basic Time Series Patterns A study on the ablty of Support Vector Regresson and Neural Networks to Forecast Basc Tme Seres Patterns Sven F. Crone, Jose Guajardo 2, and Rchard Weber 2 Lancaster Unversty, Department of Management

More information

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm Avalable onlne www.ocpr.com Journal of Chemcal and Pharmaceutcal Research, 2014, 6(7):1884-1889 Research Artcle ISSN : 0975-7384 CODEN(USA) : JCPRC5 A hybrd global optmzaton algorthm based on parallel

More information

An Interest-Oriented Network Evolution Mechanism for Online Communities

An Interest-Oriented Network Evolution Mechanism for Online Communities An Interest-Orented Network Evoluton Mechansm for Onlne Communtes Cahong Sun and Xaopng Yang School of Informaton, Renmn Unversty of Chna, Bejng 100872, P.R. Chna {chsun,yang}@ruc.edu.cn Abstract. Onlne

More information

Document image template matching based on component block list

Document image template matching based on component block list Pattern Recognton Letters 22 2001) 1033±1042 www.elsever.nl/locate/patrec Document mage template matchng based on component block lst Hanchuan Peng a,b,c, *, Fuhu Long b, Zheru Ch b, Wan-Ch Su b a Department

More information

THE APPLICATION OF DATA MINING TECHNIQUES AND MULTIPLE CLASSIFIERS TO MARKETING DECISION

THE APPLICATION OF DATA MINING TECHNIQUES AND MULTIPLE CLASSIFIERS TO MARKETING DECISION Internatonal Journal of Electronc Busness Management, Vol. 3, No. 4, pp. 30-30 (2005) 30 THE APPLICATION OF DATA MINING TECHNIQUES AND MULTIPLE CLASSIFIERS TO MARKETING DECISION Yu-Mn Chang *, Yu-Cheh

More information

ONE of the most crucial problems that every image

ONE of the most crucial problems that every image IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 23, NO. 10, OCTOBER 2014 4413 Maxmum Margn Projecton Subspace Learnng for Vsual Data Analyss Symeon Nktds, Anastasos Tefas, Member, IEEE, and Ioanns Ptas, Fellow,

More information

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression Novel Methodology of Workng Captal Management for Large Publc Constructons by Usng Fuzzy S-curve Regresson Cheng-Wu Chen, Morrs H. L. Wang and Tng-Ya Hseh Department of Cvl Engneerng, Natonal Central Unversty,

More information

Automated Network Performance Management and Monitoring via One-class Support Vector Machine

Automated Network Performance Management and Monitoring via One-class Support Vector Machine Automated Network Performance Management and Montorng va One-class Support Vector Machne R. Zhang, J. Jang, and S. Zhang Dgtal Meda & Systems Research Insttute, Unversty of Bradford, UK Abstract: In ths

More information

AUTHENTICATION OF OTTOMAN ART CALLIGRAPHERS

AUTHENTICATION OF OTTOMAN ART CALLIGRAPHERS INTERNATIONAL JOURNAL OF ELECTRONICS; MECHANICAL and MECHATRONICS ENGINEERING Vol.2 Num.2 pp.(2-22) AUTHENTICATION OF OTTOMAN ART CALLIGRAPHERS Osman N. Ucan Mustafa Istanbullu Nyaz Klc2 Ahmet Kala3 Istanbul

More information

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION Vson Mouse Saurabh Sarkar a* a Unversty of Cncnnat, Cncnnat, USA ABSTRACT The report dscusses a vson based approach towards trackng of eyes and fngers. The report descrbes the process of locatng the possble

More information

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S S C H E D A E I N F O R M A T I C A E VOLUME 0 0 On Mean Squared Error of Herarchcal Estmator Stans law Brodowsk Faculty of Physcs, Astronomy, and Appled Computer Scence, Jagellonan Unversty, Reymonta

More information

Learning from Multiple Outlooks

Learning from Multiple Outlooks Learnng from Multple Outlooks Maayan Harel Department of Electrcal Engneerng, Technon, Hafa, Israel She Mannor Department of Electrcal Engneerng, Technon, Hafa, Israel maayanga@tx.technon.ac.l she@ee.technon.ac.l

More information

Damage detection in composite laminates using coin-tap method

Damage detection in composite laminates using coin-tap method Damage detecton n composte lamnates usng con-tap method S.J. Km Korea Aerospace Research Insttute, 45 Eoeun-Dong, Youseong-Gu, 35-333 Daejeon, Republc of Korea yaeln@kar.re.kr 45 The con-tap test has the

More information

Cluster Analysis of Data Points using Partitioning and Probabilistic Model-based Algorithms

Cluster Analysis of Data Points using Partitioning and Probabilistic Model-based Algorithms Internatonal Journal of Appled Informaton Systems (IJAIS) ISSN : 2249-0868 Foundaton of Computer Scence FCS, New York, USA Volume 7 No.7, August 2014 www.jas.org Cluster Analyss of Data Ponts usng Parttonng

More information

The OC Curve of Attribute Acceptance Plans

The OC Curve of Attribute Acceptance Plans The OC Curve of Attrbute Acceptance Plans The Operatng Characterstc (OC) curve descrbes the probablty of acceptng a lot as a functon of the lot s qualty. Fgure 1 shows a typcal OC Curve. 10 8 6 4 1 3 4

More information

Multiclass sparse logistic regression for classification of multiple cancer types using gene expression data

Multiclass sparse logistic regression for classification of multiple cancer types using gene expression data Computatonal Statstcs & Data Analyss 51 (26) 1643 1655 www.elsever.com/locate/csda Multclass sparse logstc regresson for classfcaton of multple cancer types usng gene expresson data Yongda Km a,, Sunghoon

More information

Descriptive Models. Cluster Analysis. Example. General Applications of Clustering. Examples of Clustering Applications

Descriptive Models. Cluster Analysis. Example. General Applications of Clustering. Examples of Clustering Applications CMSC828G Prncples of Data Mnng Lecture #9 Today s Readng: HMS, chapter 9 Today s Lecture: Descrptve Modelng Clusterng Algorthms Descrptve Models model presents the man features of the data, a global summary

More information

Financial market forecasting using a two-step kernel learning method for the support vector regression

Financial market forecasting using a two-step kernel learning method for the support vector regression Ann Oper Res (2010) 174: 103 120 DOI 10.1007/s10479-008-0357-7 Fnancal market forecastng usng a two-step kernel learnng method for the support vector regresson L Wang J Zhu Publshed onlne: 28 May 2008

More information

FREQUENCY OF OCCURRENCE OF CERTAIN CHEMICAL CLASSES OF GSR FROM VARIOUS AMMUNITION TYPES

FREQUENCY OF OCCURRENCE OF CERTAIN CHEMICAL CLASSES OF GSR FROM VARIOUS AMMUNITION TYPES FREQUENCY OF OCCURRENCE OF CERTAIN CHEMICAL CLASSES OF GSR FROM VARIOUS AMMUNITION TYPES Zuzanna BRO EK-MUCHA, Grzegorz ZADORA, 2 Insttute of Forensc Research, Cracow, Poland 2 Faculty of Chemstry, Jagellonan

More information

A novel Method for Data Mining and Classification based on

A novel Method for Data Mining and Classification based on A novel Method for Data Mnng and Classfcaton based on Ensemble Learnng 1 1, Frst Author Nejang Normal Unversty;Schuan Nejang 641112,Chna, E-mal: lhan-gege@126.com Abstract Data mnng has been attached great

More information

IMPACT ANALYSIS OF A CELLULAR PHONE

IMPACT ANALYSIS OF A CELLULAR PHONE 4 th ASA & μeta Internatonal Conference IMPACT AALYSIS OF A CELLULAR PHOE We Lu, 2 Hongy L Bejng FEAonlne Engneerng Co.,Ltd. Bejng, Chna ABSTRACT Drop test smulaton plays an mportant role n nvestgatng

More information

Adaptive Fractal Image Coding in the Frequency Domain

Adaptive Fractal Image Coding in the Frequency Domain PROCEEDINGS OF INTERNATIONAL WORKSHOP ON IMAGE PROCESSING: THEORY, METHODOLOGY, SYSTEMS AND APPLICATIONS 2-22 JUNE,1994 BUDAPEST,HUNGARY Adaptve Fractal Image Codng n the Frequency Doman K AI UWE BARTHEL

More information

Eye Center Localization on a Facial Image Based on Multi-Block Local Binary Patterns

Eye Center Localization on a Facial Image Based on Multi-Block Local Binary Patterns Eye Center Localzaton on a Facal Image Based on Mult-Bloc Local Bnary Patterns Anatoly tn, Vladmr Khryashchev, Olga Stepanova Yaroslavl State Unversty Yaroslavl, Russa anatolyntnyar@gmal.com, vhr@yandex.ru,

More information

Realistic Image Synthesis

Realistic Image Synthesis Realstc Image Synthess - Combned Samplng and Path Tracng - Phlpp Slusallek Karol Myszkowsk Vncent Pegoraro Overvew: Today Combned Samplng (Multple Importance Samplng) Renderng and Measurng Equaton Random

More information

Inter-Ing 2007. INTERDISCIPLINARITY IN ENGINEERING SCIENTIFIC INTERNATIONAL CONFERENCE, TG. MUREŞ ROMÂNIA, 15-16 November 2007.

Inter-Ing 2007. INTERDISCIPLINARITY IN ENGINEERING SCIENTIFIC INTERNATIONAL CONFERENCE, TG. MUREŞ ROMÂNIA, 15-16 November 2007. Inter-Ing 2007 INTERDISCIPLINARITY IN ENGINEERING SCIENTIFIC INTERNATIONAL CONFERENCE, TG. MUREŞ ROMÂNIA, 15-16 November 2007. UNCERTAINTY REGION SIMULATION FOR A SERIAL ROBOT STRUCTURE MARIUS SEBASTIAN

More information

The Network flow Motoring System based on Particle Swarm Optimized

The Network flow Motoring System based on Particle Swarm Optimized The Network flow Motorng System based on Partcle Swarm Optmzed Neural Network Adult Educaton College, Hebe Unversty of Archtecture, Zhangjakou Hebe 075000, Chna Abstract The compatblty of the commercal

More information

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Luby s Alg. for Maximal Independent Sets using Pairwise Independence Lecture Notes for Randomzed Algorthms Luby s Alg. for Maxmal Independent Sets usng Parwse Independence Last Updated by Erc Vgoda on February, 006 8. Maxmal Independent Sets For a graph G = (V, E), an ndependent

More information

8 Algorithm for Binary Searching in Trees

8 Algorithm for Binary Searching in Trees 8 Algorthm for Bnary Searchng n Trees In ths secton we present our algorthm for bnary searchng n trees. A crucal observaton employed by the algorthm s that ths problem can be effcently solved when the

More information

New Approaches to Support Vector Ordinal Regression

New Approaches to Support Vector Ordinal Regression New Approaches to Support Vector Ordnal Regresson We Chu chuwe@gatsby.ucl.ac.uk Gatsby Computatonal Neuroscence Unt, Unversty College London, London, WCN 3AR, UK S. Sathya Keerth selvarak@yahoo-nc.com

More information

Vehicle Detection and Tracking in Video from Moving Airborne Platform

Vehicle Detection and Tracking in Video from Moving Airborne Platform Journal of Computatonal Informaton Systems 10: 12 (2014) 4965 4972 Avalable at http://www.jofcs.com Vehcle Detecton and Trackng n Vdeo from Movng Arborne Platform Lye ZHANG 1,2,, Hua WANG 3, L LI 2 1 School

More information

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT Toshhko Oda (1), Kochro Iwaoka (2) (1), (2) Infrastructure Systems Busness Unt, Panasonc System Networks Co., Ltd. Saedo-cho

More information

Research Article Integrated Model of Multiple Kernel Learning and Differential Evolution for EUR/USD Trading

Research Article Integrated Model of Multiple Kernel Learning and Differential Evolution for EUR/USD Trading Hndaw Publshng Corporaton e Scentfc World Journal, Artcle ID 914641, 12 pages http://dx.do.org/10.1155/2014/914641 Research Artcle Integrated Model of Multple Kernel Learnng and Dfferental Evoluton for

More information

Assessing Student Learning Through Keyword Density Analysis of Online Class Messages

Assessing Student Learning Through Keyword Density Analysis of Online Class Messages Assessng Student Learnng Through Keyword Densty Analyss of Onlne Class Messages Xn Chen New Jersey Insttute of Technology xc7@njt.edu Brook Wu New Jersey Insttute of Technology wu@njt.edu ABSTRACT Ths

More information

A Fast Incremental Spectral Clustering for Large Data Sets

A Fast Incremental Spectral Clustering for Large Data Sets 2011 12th Internatonal Conference on Parallel and Dstrbuted Computng, Applcatons and Technologes A Fast Incremental Spectral Clusterng for Large Data Sets Tengteng Kong 1,YeTan 1, Hong Shen 1,2 1 School

More information

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services An Evaluaton of the Extended Logstc, Smple Logstc, and Gompertz Models for Forecastng Short Lfecycle Products and Servces Charles V. Trappey a,1, Hsn-yng Wu b a Professor (Management Scence), Natonal Chao

More information

+ + + - - This circuit than can be reduced to a planar circuit

+ + + - - This circuit than can be reduced to a planar circuit MeshCurrent Method The meshcurrent s analog of the nodeoltage method. We sole for a new set of arables, mesh currents, that automatcally satsfy KCLs. As such, meshcurrent method reduces crcut soluton to

More information

Loop Parallelization

Loop Parallelization - - Loop Parallelzaton C-52 Complaton steps: nested loops operatng on arrays, sequentell executon of teraton space DECLARE B[..,..+] FOR I :=.. FOR J :=.. I B[I,J] := B[I-,J]+B[I-,J-] ED FOR ED FOR analyze

More information

1 Example 1: Axis-aligned rectangles

1 Example 1: Axis-aligned rectangles COS 511: Theoretcal Machne Learnng Lecturer: Rob Schapre Lecture # 6 Scrbe: Aaron Schld February 21, 2013 Last class, we dscussed an analogue for Occam s Razor for nfnte hypothess spaces that, n conjuncton

More information

Vehicle Detection, Classification and Position Estimation based on Monocular Video Data during Night-time

Vehicle Detection, Classification and Position Estimation based on Monocular Video Data during Night-time Vehcle Detecton, Classfcaton and Poston Estmaton based on Monocular Vdeo Data durng Nght-tme Jonas Frl, Marko H. Hoerter, Martn Lauer and Chrstoph Stller Keywords: Automotve Lghtng, Lght-based Drver Assstance,

More information

Texture Analysis and Modified Level Set Method for Automatic Detection of Bone Boundaries in Hand Radiographs

Texture Analysis and Modified Level Set Method for Automatic Detection of Bone Boundaries in Hand Radiographs Texture Analyss and Modfed Level Set Method for Automatc Detecton of Bone Boundares n Hand Radographs Syaful Anam Graduate School of Scence and Engneerng Yamaguch Unversty Yamaguch, Japan Department of

More information

STANDING WAVE TUBE TECHNIQUES FOR MEASURING THE NORMAL INCIDENCE ABSORPTION COEFFICIENT: COMPARISON OF DIFFERENT EXPERIMENTAL SETUPS.

STANDING WAVE TUBE TECHNIQUES FOR MEASURING THE NORMAL INCIDENCE ABSORPTION COEFFICIENT: COMPARISON OF DIFFERENT EXPERIMENTAL SETUPS. STADIG WAVE TUBE TECHIQUES FOR MEASURIG THE ORMAL ICIDECE ABSORPTIO COEFFICIET: COMPARISO OF DIFFERET EXPERIMETAL SETUPS. Angelo Farna (*), Patrzo Faust (**) (*) Dpart. d Ing. Industrale, Unverstà d Parma,

More information

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ). REVIEW OF RISK MANAGEMENT CONCEPTS LOSS DISTRIBUTIONS AND INSURANCE Loss and nsurance: When someone s subject to the rsk of ncurrng a fnancal loss, the loss s generally modeled usng a random varable or

More information

A Suspect Vehicle Tracking System Based on Video

A Suspect Vehicle Tracking System Based on Video 3rd Internatonal Conference on Multmeda Technology ICMT 2013) A Suspect Vehcle Trackng System Based on Vdeo Yad Chen 1, Tuo Wang Abstract. Vdeo survellance systems are wdely used n securty feld. The large

More information

Calculation of Sampling Weights

Calculation of Sampling Weights Perre Foy Statstcs Canada 4 Calculaton of Samplng Weghts 4.1 OVERVIEW The basc sample desgn used n TIMSS Populatons 1 and 2 was a two-stage stratfed cluster desgn. 1 The frst stage conssted of a sample

More information

Improved SVM in Cloud Computing Information Mining

Improved SVM in Cloud Computing Information Mining Internatonal Journal of Grd Dstrbuton Computng Vol.8, No.1 (015), pp.33-40 http://dx.do.org/10.1457/jgdc.015.8.1.04 Improved n Cloud Computng Informaton Mnng Lvshuhong (ZhengDe polytechnc college JangSu

More information

where the coordinates are related to those in the old frame as follows.

where the coordinates are related to those in the old frame as follows. Chapter 2 - Cartesan Vectors and Tensors: Ther Algebra Defnton of a vector Examples of vectors Scalar multplcaton Addton of vectors coplanar vectors Unt vectors A bass of non-coplanar vectors Scalar product

More information

RESEARCH ON DUAL-SHAKER SINE VIBRATION CONTROL. Yaoqi FENG 1, Hanping QIU 1. China Academy of Space Technology (CAST) yaoqi.feng@yahoo.

RESEARCH ON DUAL-SHAKER SINE VIBRATION CONTROL. Yaoqi FENG 1, Hanping QIU 1. China Academy of Space Technology (CAST) yaoqi.feng@yahoo. ICSV4 Carns Australa 9- July, 007 RESEARCH ON DUAL-SHAKER SINE VIBRATION CONTROL Yaoq FENG, Hanpng QIU Dynamc Test Laboratory, BISEE Chna Academy of Space Technology (CAST) yaoq.feng@yahoo.com Abstract

More information

Project Networks With Mixed-Time Constraints

Project Networks With Mixed-Time Constraints Project Networs Wth Mxed-Tme Constrants L Caccetta and B Wattananon Western Australan Centre of Excellence n Industral Optmsaton (WACEIO) Curtn Unversty of Technology GPO Box U1987 Perth Western Australa

More information

Implementation of Deutsch's Algorithm Using Mathcad

Implementation of Deutsch's Algorithm Using Mathcad Implementaton of Deutsch's Algorthm Usng Mathcad Frank Roux The followng s a Mathcad mplementaton of Davd Deutsch's quantum computer prototype as presented on pages - n "Machnes, Logc and Quantum Physcs"

More information

BERNSTEIN POLYNOMIALS

BERNSTEIN POLYNOMIALS On-Lne Geometrc Modelng Notes BERNSTEIN POLYNOMIALS Kenneth I. Joy Vsualzaton and Graphcs Research Group Department of Computer Scence Unversty of Calforna, Davs Overvew Polynomals are ncredbly useful

More information

Development of an intelligent system for tool wear monitoring applying neural networks

Development of an intelligent system for tool wear monitoring applying neural networks of Achevements n Materals and Manufacturng Engneerng VOLUME 14 ISSUE 1-2 January-February 2006 Development of an ntellgent system for tool wear montorng applyng neural networks A. Antć a, J. Hodolč a,

More information

DEFINING %COMPLETE IN MICROSOFT PROJECT

DEFINING %COMPLETE IN MICROSOFT PROJECT CelersSystems DEFINING %COMPLETE IN MICROSOFT PROJECT PREPARED BY James E Aksel, PMP, PMI-SP, MVP For Addtonal Informaton about Earned Value Management Systems and reportng, please contact: CelersSystems,

More information

Logistic Regression. Steve Kroon

Logistic Regression. Steve Kroon Logstc Regresson Steve Kroon Course notes sectons: 24.3-24.4 Dsclamer: these notes do not explctly ndcate whether values are vectors or scalars, but expects the reader to dscern ths from the context. Scenaro

More information

Offline Verification of Hand Written Signature using Adaptive Resonance Theory Net (Type-1)

Offline Verification of Hand Written Signature using Adaptive Resonance Theory Net (Type-1) Internatonal Journal of Sgnal Processng Systems Vol, No June 203 Offlne Verfcaton of Hand Wrtten Sgnature usng Adaptve Resonance Theory Net (Type-) Trtharaj Dash Veer Surendra Sa Unversty of Technology,

More information

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy Fnancal Tme Seres Analyss Patrck McSharry patrck@mcsharry.net www.mcsharry.net Trnty Term 2014 Mathematcal Insttute Unversty of Oxford Course outlne 1. Data analyss, probablty, correlatons, vsualsaton

More information

Parallel Numerical Simulation of Visual Neurons for Analysis of Optical Illusion

Parallel Numerical Simulation of Visual Neurons for Analysis of Optical Illusion 212 Thrd Internatonal Conference on Networkng and Computng Parallel Numercal Smulaton of Vsual Neurons for Analyss of Optcal Illuson Akra Egashra, Shunj Satoh, Hdetsugu Ire and Tsutomu Yoshnaga Graduate

More information

Recurrence. 1 Definitions and main statements

Recurrence. 1 Definitions and main statements Recurrence 1 Defntons and man statements Let X n, n = 0, 1, 2,... be a MC wth the state space S = (1, 2,...), transton probabltes p j = P {X n+1 = j X n = }, and the transton matrx P = (p j ),j S def.

More information

How To Create An Emoton Recognzer

How To Create An Emoton Recognzer Recognzng Low/Hgh Anger n Speech for Call Centers FU-MING LEE*, LI-HUA LI, RU-YI HUANG Department of Informaton Management Chaoyang Unversty of Technology 168 Jfong E. Rd., Wufong Townshp Tachung County,

More information