Study On Uyghur Off-Line Handwritten Signature Verification Based On Local Features

Biometrics technology has gradually replaced the traditional password encryption verification method.Offline handwritten signature verification is one of the important fields in biometric verification.Offline handwritten signature verification has been applied in many fields,such as finance,justice,etc.Offline handwritten signature is stored in the form of static image,which only has static features.Therefore,how to extract effective signature image features is very important in offline handwritten signature verification research.In this paper,the verification of handwritten signature from Uyghur and offline handwritten signature in CEDAR database is studied.By analyzing the characteristics of the signature samples,the specificity between the signatures is mainly reflected in the local differences in the samples.Therefore,features based on edge information and block theory is proposed,and feature fusion is used to improve signature verification accuracy.The main work of this paper is as follows:1)This paper introduces and summarizes the relevant concepts and research status of offline handwritten signature verification technology,and analyzes the difficulties in the research of offline handwritten signature verification.2)The signature sample images are preprocessed.Bilinear interpolation,weighted average method,Otsu and bilateral filtering were used for size normalization,gray scale,binarization and smooth denoising.3)A fusion texture feature based on block theory is proposed.After the signature is processed,the MB-LBP and LPQ features are extracted from the image in each block,and all extracted texture features are fused into high-dimensional texture features based block.Support vector machine(SVM)and random forest classifier are used to identify the signature.The total accuracy is 96.06% from Uyghur signature database,and 97.04% from CEDAR database.4)A high dimensional feature based on edge information is proposed.After obtaining the edge image of the signature sample,the SURF and ORB feature points are extracted to establish a modified visual word bag model based on the edge information.The modified visual word bag model was fused with edge direction histogram to form a high-dimensional feature based on edge information.SVM and RF were used for signature verification respectively,and the Overall Right Rate of signature verification for Uyghur was 93.69%,and the Overall Right Rate of signature verification for CEDAR database was 96.17%.By the experiment of signature from the Uyghur handwriting signature database and CEDAR database,the highest Overall Right Rate is 96.06% and 97.04%,respectively.Therefore,the method proposed in this paper has good accuracy in offline handwritten signature verification.