Artificial Impostor Profiling for Keystroke Dynamics on a Mobile Handset |
Keystroke Analysis is a biometric approach that utilises the typing characteristics of a user to perform identity authentication, and has two key advantages in a mobile context – the necessary authentication hardware (i.e. the keypad is already present) and the technique can operate transparently. Although studies have proved the feasibility of such an approach on a mobile handset, a failing exists in the practical deployment of the system. Classification is performed by neural networks that are trained using both the authorised users samples and impostors as a means of comparison. However, in the real world, the availability and suitability of impostor samples will be limited. This paper proposes a means of artificially creating impostor data directly based upon samples from the authorised user in order to provide optimally configured classification engines. These artificial impostor approaches have not only solved the availability issue but have improved the system performance (in comparison to the traditional approach) by up to 25%.
Lecomte J, Clarke NL, Furnell SM