On-line Handwriting Signature Verification Technology Based On Feature Optimal Selection And Neural Network Classifier

Today, with the rapid development of information and internet, information security is extraordinarily important, especially, personal identification technology has significant application. Because signature verification technology has stability, uniqueness and convenience, it is applied more and more broadly. As handwriting signature is typical of ease acquisition, moreover, its major feature is apparent, stability and good divisibility. Signature verification system has friendship, convenience and more and more broadly application, so signature verification is a potential personal identification technology.Signature verification system use the touch panel and touch panel controller ADS7846 to collect signature data (include signature horizontal coordinate, vertical coordinate, pressure information). We wipe off noise and eliminate effective-zero but save the location of these effective-zero. Useless or interferential information are eliminated during preprocessing. The research of this paper mainly includes two aspects; firstly, we carry out a feature election on preprocessed signature. This paper respectively process feature extraction using only genuine signature, feature extraction using both genuine and forgery signature, common feature subset extraction experiments. The signature verification system will have faster response time and less memory size, and the best feature subset are confirmed by system verification performance. Second, Because neural network has abilities of self-study, self-adapt and better contain-error, so the paper designs an ameliorative BP network. During the design BP classifier process, we use synthesized algorithm which makes learning time short, convergence fast. At the same time, we deeply research the ascertainment of the node of latent-layer, the election of study-gene η and inertial-item α in BP neural network and so on. The result of experiment proves that all kinds of genuine signatures and forged signatures input trained BP neural networks acquired good verification result.