Neural networks for signature classification and identity verification

A digitizing tablet was employed to collect handwritten signatures with five quantities recorded, namely horizontal and vertical pen tip position, pen tip pressure, and pen azimuth and altitude angles. Cluster analysis was applied to segment the feature space into sub-regions of “similar” signatures to facilitate the classification and verification tasks. Both problems were solved with the use of neural classifiers. Only the hidden features, namely those not visible in signature images, were used for verification purposes. A two-layer sigmoidal perceptron was used to approximate the classification function. An adaptive radial-basis network called the RCE network was also implemented to compare capabilities of both networks.