Line Chinese Signature Verification Technology

Signature is one of the biological features used abroad to verify identity. Because of the significant advantages of signature verification as a way of personal identification, it has a wide perspective in applications in electronic-bank, military affairs, electronic-business, communication, office automation and engineering files. The research into signature verification is indeed significant both to practical application and the development of science.First, the application background, the development history and the researches at home and abroad of handwritten signature verification techniques are briefly introduced, The difficulties and main research methods of the off-line Chinese signature verification problem and are analyzed. Then, the key techniques including image preprocessing, feature extraction and verification of off-line signature have been studied in depth in this thesis. The main research work including:Most of the methods used for preprocessing of the signature image are ripe algorithms for off-line Chinese signature verification. This part introduces the image filtering, binarization, margin disposing, slant correction, horizontal compress, and normalization.The signatures' static shape feature and pseudo dynamic features have been extracted in feature extraction. The emphasis of shape feature is the pseudo-Zernike moment with rotation, scale and translation invariance and anti-noise character. The pseudo dynamic features have original gray feature, high gray feature and gray histogram feature.Combine the advantage of Genetic Algorithm and Wavelet Neural Network, this paper proposes two models for Wavelet Neural Network based on Genetic Algorithm. In the first model, instead of gradient descent algorithm in WNN, GA is used to optimize the parameters for WNN. In the second model, GA is used to locate the initial parameters for WNN firstly and then the gradient descent algorithm is employed to optimize these parameters. Then the proposed two GA-WNN models are used to the off-line Chinese signature verification. The experimental results prove that the first model can reduce the operation, and the second model can descend the average rate of false recognition to 9.2% compared with BPNN and WNN.When there are a few genuine signature samples and less or non forgery signature samples, considering the relative stabilization of genuine signatures and the stochastic levity of forgery signatures, this paper proposes a method based on few reference genuine signatures. After we project the testing signatures data to the principal component vectors of these reference genuine signatures, the new transformed signatures data is not depend on any forgery signature. By this way, we can get 11.78% average rate of false recognition when there are only 4 or 5 genuine signatures and on one forgery.