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

As a kind of identification technology in the field of biometric recognition,handwritten signature verification has been widely accepted and applied in finance,legal,commerce and so on.Currently,handwritten signature verification technology based on English,Arabic,and Chinese have been got more mature researches and results,but Uyghur handwritten signature verification is still in a starting stage in the field.Therefore,it has great practical application and value to conduct further research to Uyghur handwritten signature verification and to make up and complete the signature verification technology of ethnic minorities in our country.This paper was mainly studied and analyzed for offline Uyghur handwriting signature verification technology.The research work includes signature sample acquisition and preprocessing,feature extraction,classification and verification of three parts.In the preprocessing stage the noise and interference signals on the signature image were evaluated by the graying,binarization,smoothing and normalization of the signature image.In the stage of feature extraction,according to the writing style and nature of Uyghur handwritten signatures,four different scans of each signature sample image were presented with a 16-dimensional direction feature.Secondly,based on the feature extraction method of directional feature,an improved 48-dimensional direction feature was proposed based on the black pixel in six different directions.Finally,based on the feature parameters such as energy,entropy,moment of inertia and local smoothness of gray level co-occurrence matrix,the feature of weighted was used to extract the fusion feature and the optimal weight for Uyghur handwritten signature is determined.In the classification phase of the signature image,the two kinds of directional features proposed in this paper were used to verify the signatures by using three distance classifiers such as Euclidean distance,chi-square distance and Manhattan distance.For the gray-level co-occurrence matrix weighted fusion feature,the BP neural network was used to identify the signature.In the experiment,from the Uygur handwritten signature sample library selected 15 individuals(20 original signature sample / per person + 20 simple forgery signature sample / per person +20 skilled forgery signature sample / per person)900 handwritten signature samples.Finally,the highest signature verification rate obtained by the three signature identification methods proposed in this paper was 88.61%,96% and 91.78% respectively.