Biometric identification for dynamic signature verification using time delay neural networks

Many biometric identification methods are available for verifying an individual's identity and the handwritten signature is one with a long history. This well established custom is used for identification because the handwritten signature is unique to each individual. While each individual has a unique signature there are a number of variations that occur with each signing. The handwritten signature is a behavior that is learned and changes over the life time of an individual. Even with these variations the ability of a human to recognize a handwritten signature is very robust, yet for a computer system to accomplish this same task has not always been as successful.; Now with the passage of the e-commerce act of 2000, the electronic signature carries the same weight under the law as their handwritten counterparts. This law, which takes effect on March 1, 2001 declares the validity of electronic signatures for world wide commerce. The law defines a Digital Signature Standard (DSS) that supports all forms of electronic signature technologies. This law is causing the acceleration and development of biometric verification methods and dynamic signature verification is one of the technologies to address this requirement.; The requirement for signature verification is to accept and ensure the identity of an individual and reject any forgeries. The trend in handwritten signature verification systems is towards more advanced verification algorithms and the development of higher resolution input devices.; This dissertation reviews the current state of the art in signature verification and implements a dynamic signature verification system using the standard components available on a personal computer today. Specifically, this dissertation develops a dynamic signature collection system using the GNOME/GTK/X libraries and a dynamic signature verification method using time delay neural networks. Experimental results from a signature database shows that the dynamic signature verification method using time delay neural networks is effective and that biometric: identification using a low resolution input device is possible.