Off-line Signature Verification Based On Dual Bag Of Visual Words Model

The traditional off-line handwritten signature identification is mainly based on the artificial method,which is inadequate in both efficiency and accuracy rate.With the rapid development of computer technology,many researchers use computer technology to carry out off-line signature identification work,and there are still many problems and deficiencies.Chinese off-line signature identification work is still very few,especially.As a result,study further is necessary.This paper proposes a method of off-line signature identification based on double word bag model,and the main research contents are as follows:(1)For traditional off-line signature preprocessing method,such as signature image binary,smoothing,denoising,compression signature,skeleton extraction and other methods inevitably lead to the loss of signature image information,this paper proposes a preprocessing scheme based on image mask clipping,which maximum preserves the effective information of the original signature image.(2)Noticed in the most signature image,the features are only in in the stroke area,we proposes a PULBP(Partial-uniform LBP)feature based on the stroke region,which effectively solves the problem that effective information occupies a low proportion of the traditional ULBP operator in the signature image feature extraction process;(3)In this paper,a method based on double vision word bag model is proposed,which integrates multiple features organically to improve the robustness of off-line handwritten signature authentication algorithm.This method can solve the problem that in the step of traditional feature fusion,in which the direct stitching method is used for the most of time,may lead to the change of feature weights implicitly.(4)Given the lack of research on Chinese off-line signature identification,this algorithm takes into account the characteristics of Chinese signature,and the experimental results show that this method has substantial robustness to Chinese signature identification.(5)In this paper,a collection of Chinese signature is carried out,and a total of 40 signature images are collected,as well as the skillful imitation signatures of the other.For each signature,we collected 30 authentic signatures and 30 skilled pseudo signatures.The results of this method are verified experimentally in two English signature data sets and the Chinese data set by the author,and the experimental result shows that the proposed method has better robustness and adaptability,and compared with other stateof-the-art,it has higher accuracy and higher equal error rate in signature authenticity recognition.