Neural based signature analysis with application to banking

This paper presents an algorithm for signature analysis using artificial neural networks (ANN). A new technique, based upon the adaptation of ANN to recognize signatures of connected type patterns, is introduced. The proposed technique separates the signature from noise. It worthy to notice that this technique rejects any signature has not trained before. The introduced ANN was trained with backpropagation with momentum and an ANN with 400 input, 40 hidden layer and 5 neurons in the output layer gives good performance with compared with other network structures. The two layers employ log-sigmoid transfer functions. Errors are minimized through the use of a normalizing procedure. The introduced algorithm can be applied to locate faults in digital systems, security systems, banking, and other disciplines. Moreover, it has the advantages of simplicity, accuracy, fast and easy implementation