Study On Uyghur Off-line Handwritten Signature Veriifcation

As a kind of behavioral feature, handwritten signature is used in identityauthentication, and it has been widely applied to economics, financial, legal industriesetc. With the rapid development of economics and increasing exchanges betweenother countries, it is required sign the handwritten signature on the documents such asagreements, contracts, proposal books, checks and so on. It will bring out very seriousconsequences; even destroy the normal operation of whole financial system if thesignature was imitated. Therefore, it has significant value and practical meaning tocarry out reliable, effective and rapid signature verification method.The exploratory studies carried for offline Uyghur handwritten signatureverification first time in this paper. The main idea of the paper is indicated as thefollowing: firstly, preprocessing is conducted to the original signature image. Then,features are extracted from them. Finally, it is confirmed by using classifiers thesignature’s true or false after matching features between each feature vector in thefeature database. In the preprocessing step, it is necessary to normalize the signatureimage, because the different size and location of the signatures from different people.Then, the differential impact on the background is considered, denoising andsmoothing work is conducted for careless writing handwriting and other factors.In the feature extraction part, it is separately extracted1dimensional baselinefeatures，2dimensional upper and lower boundary features, and128dimensional localcentral point features after segmenting the signature image vertical and the horizontaldirection. In the verification part, Canberra distance, block distance, Dice coefficientand KNN classifier are used. The extracted features are classified after measuringthe distance via distance classifier. Due to the same FAR and FRR in the distanceclassifier, the threshold value is different in the circumstance of same features andsame classifiers are used different peoples’ signatures, so the threshold selecting method is used in this paper. The comparative analysis which Canberra distance,block distance and Dice coefficient effects on the verification rates using differentfeatures is also studied here as well. It is achieved95%and97.5%of highest overallaccuracy rates (ORR) using2types of features and3kinds of classifiers on the300Uyghur handwritten signatures signed by different5persons (each person asked tosign20samples of original signature, simple forgery signatures and skilled forgerysignatures respectively), correspondingly, obtained value of FAR and FRR separatelyis0%and10%from using the base line features, upper and lower boundary features,and5%and0%from the local central point features. The systems ORR increased to99%when the128dimensional local central point feature and baseline featurescombined together. The experimental results indicated that it is an effectiveverification method for Uyghur handwritten signature.