Wavelet Based Approach for Off Line Handwritten Signatures Verification

Abstract

Though a lot of research has been undertaken in thearea of handwritten signatures verification, the recognition rates stillneed to be improved. The low recognition rates are largely attributedto the fact thatthere are intra personal variations in an individual’ssignature. In this paper a wavelet–based off–line signatureverification system is proposed. The proposed system extracts the lowfrequency components and also the high frequency components whichare also known as approximation coefficients and detail coefficientsrespectively in wavelet terminology. The high frequency componentsrepresent the fast changing parts of the signal and the low frequencycomponents represent the less varying or smooth parts of the signal.The handwritten signature images written on the papers are scannedinto the machine and stored in jpeg format. The signatures images arebinarized and bounding rectangles are put covering only the signaturearea. The bounded signatures are normalized using the Bicubicinterpolation method and are thinned. Since the signatures of the sameperson though not identical, but definitely exhibit some form ofconsistency or stability, hence regularity analysis has been done usinga regular wavelet like Daubechies wavelet transform on thepreprocessed handwritten signature images. The decomposition isdone for six levels on the signature images which are in jpeg formatand the principal component analysis is done. The principalcomponents are chosen according to the ‘kais’ rule. The principalcomponent vector is used to train the standard K-NN classifier andclassification is done. The results of the proposed method are quitesatisfactory in case of random forgeries. Certainly for the skilledforgeries, the method needs to be improved.