Feature Selection on Handwriting Biometrics: Security Aspects of Artificial Forgeries

A lot of improvements were introduced lately in order to increase the verification performance of biometric user authentication systems. One method, besides many others, is the selection of specific features for each user during the verification process. In this paper we present a security analysis of a user specific bit mask vector, which was originally introduced to improve verification performance on a Biometric Hash algorithm for dynamic handwriting. Therefore, we use a reverse engineering attack method to generate artificial handwriting data and calculate error rates to examine the impact on the verification performance. Our goal is to study the effect of a feature selection by a mask vector on artificial data in comparison to genuine handwriting data. Our first experimental results show an average decrease of the equal error rate, generate by the artificial data, by approx. 64%. In comparison, equal error rates of random attacks, using verification data of another user, decreases by an average