Study On The On-Line Signature Handwriting Verification Techniques Based On Neural Network

With the rapidly development of computer information technology and highly desire of information security. As one of the biometrics technology, on-line signature handwriting verification technology is getting more and more attention because of its with out touch, convenience used and long tradition in many common commercial fields.Signature verification system uses the touch panel to collect signature data like horizontal coordinate, vertical coordinate, pressure information. Preprocessing with wiping off noise and so on to eliminate useless or interferential information. The thesis focuses on the studying of the extraction of the features and classifying the genuine and forgery signatures by Neural Network. In the features extraction aspect, gets the local and whole parameter features like velocity, acceleration. Carries out a feature election on proportioned parameter features. Filter twice the features in the considering of the distance of one feature in the genuine and forgery signatures and one feature in the genuine signatures. With many experiments of the two-kind of feature election methods, proves that 24 elected feature are more representative to the signatures.Designs BP Neural Network classifier to classify the genuine and forgery signatures. Trains the feature samples with genuine and forgery feature samples. Saves the convergent weight value. Input the feature samples into the net. Judges the signature is genuine or forgery by the comparisons of the threshold and the export. Regulates the suitable method of initializing the weight values and standard of the ending of learning. Improves the Sigmoid function of the net nodes. Imports the method of self-adjusting study-gene and inertial-item to make the convergence fast. According to the feature samples, deeply researches the ascertainment of the node of latent-layer. The thesis tests the signature verification system using many kinds of signatures. The result of experiment proves that signature verification system using dynamic features and BP Neural Network acquired good verification result.