Dynamic Online Signatures Recognition System Using a Novel Signature-based Normalized Features String and Mlp Neural Network

In a conventional authentication process, every one may have a signature prototype that plays the role of valid benchmark. To recognize the online signature patterns, we proposed a dynamic recognition system based on a novel signature-based normalized features string (SNFS) as extracted features, and a multilayer perceptron (MLP) neural network as the classifier. We showed that the proposed SNFS is completely rotation, scale, and shift invariant feature which is on the other hand correlated mostly with speed and shape of the input signatures which are two important parameters of online signatures. As time goes by, signature prototype of persons may change. The classifier not only should recognize the signature prototype, but also must train and update itself based on the little changes in a recognized signature, therefore we used multi layer perceptron (MLP) neural network as a classifier. After each recognition, MLP is re-trained and its weights are modified within several learning epochs. The proposed algorithm was applied to a set of 80 signature patterns of 10 IRANIAN JOURNAL OF ENGINEERING SCIENCES VOL. 1, NO.1, DECEMBER 2007