Signature verification system on neuro-computer

A signature verification system using a neuro-computer is presented. The neuro-computer executes the backpropagation learning algorithm for multilayer neural networks faster than a super computer. The system uses signals obtained from pen pressure and writing speed. A method for generating deformed signature patterns is also presented. The learning of about 500 patterns is converged within several seconds using four-layer networks (768-119-61-1). The evaluation result based on 70 persons' signatures is also presented. The reject ratio (false rejection) is 7.97% and the error ratio (false acceptance) is 0.61%.<
>