(IJCSIS) International Journal of Computer Science and Information Security,Vol.
2D strings. We have proposed an offline signature retrievalmodel based on global features.The main contribution of this paper is that, we have proposedoff-line handwritten signature retrieval using curvelettransform, In retrieval phases Canberra distance measure isused. The experimental results of proposed method weresatisfactory and found that it had better results compare withrelated works. The rest of paper is organized as follows: Insection II, discusses the feature extraction phase. The signatureretrieval is presented in section III. In section IV, theexperimental results and finally section V concludes thework.II.FEATURE EXTRACTION PHASEThe major task of feature extraction is to reduce image data tomuch smaller amount of data which represents the importantcharacteristic of the image. In signature retrieval, edgeinformation is very important in characterizing signature properties. Therefore we proposed to use the curvelettransform. The performance of the system is compared withstandard discrete wavelet transform which capturesinformation in only three directions.
A.Discrete Wavelet Transform
The multi resolution wavelet transform decomposes asignal into low pass and high pass information. The low passinformation represents a smoothed version and the main bodyof the original data. The high pass information represents dataof sharper variations and details. Discrete Wavelet Transformdecomposes the image into four sub-images when one level of decomposing is used. One of these sub-images is a smoothedversion of the original image corresponding to the low passinformation and the other three ones are high pass informationthat represents the horizontal, vertical and diagonal edges of theimage respectively. When two images are similar, their difference would be existed in high-frequency information. ADWT with N decomposition levels has 3N+1 frequency bandswith 3N high-frequency bands , . The impulse responsesassociated with 2-D discrete wavelet transform are illustrated inFig. 1 as gray-scale image.
Fig. 1.Impulse response of 0
Recently, Candµes and Donoho developed a newmultiscale transform which they called the curvelet transform.Motivated by the needs of image analysis, it was neverthelessfirst proposed in the context of objects f(x1, x2) defined on thecontinuum plane (x1, x2) € R
.The transform was designed to represent edges and other singularities along curves much more efficiently thantraditional transforms, i.e. using many fewer coefficients for agiven accuracy of reconstruction. Roughly speaking, torepresent an edge to squared error 1/N requires 1/N wavelets.The curvelet transform, like the wavelet transform, is amultiscale transform, with frame elements indexed by scale andlocation parameters. Unlike the wavelet transform, it hasdirectional parameters, and the curvelet pyramid containselements with a very high degree of directional specificity. Inaddition, the curvelet transform is based on a certainanisotropic scaling principle which is quite different from theisotropic scaling of wavelets. The elements obey a specialscaling law, where the length of the support of a frameelements and the width of the support are linked by the relationwidth ≈ length
.see details in .
C.Feature Database Creation
To construct the feature vectors of each handwrittensignature in the database using DWT and curvelet transformrespectively. The Energy and Standard Deviation (STD) werecomputed separately on each sub band and the feature vector was formed using these two parameter values. The Energy
and Standard Deviation
sub band iscomputed as follows
M i N jk k
jiW N M E
( )( )
N iM jk k k
jiW N M
is the size of wavelet decomposed sub band, and
is the mean of the
sub band. Theresulting feature vector using energy and standard deviation are
E E E f
σ σ σ
respectively. So combinedfeature vector is
E E E f
σ σ σ
(3)III. OFFLINE H
PHASEThere are several ways to work out the distance betweentwo points in multidimensional space. The most commonlyused is the Canberra distance measure. It can be considered theshortest distance between two points. We have used Canberradistance metric as similarity measure. If x and y are the featurevectors of the database and query signature, respectively, x andy have dimension d, then the Canberra distance is given byCanb (x, y)
y x y x