A comparative analysis on the performance of static handwritten verification systems on realistic scenarios

Abstract

An analysis on realistic conditions for static handwritten signature verification systems is presented. Evaluation conditions are studied so to approximate as accurately as possible the conditions of a real-world operational scenario. Also, the problem of modeling realistic attacks by using databases containing forgeries is considered in the analysis. To do so, a database collected with different levels of knowledge and training is used. Performance evaluation results with these restrictions are given over several databases. Results show the effect of the amount of training and the information available to the forger on the attack resistance of some algorithms.