Enhanced template update

With the increasing number of activities being performed using computers, there is an ever growing need for advanced authentication mechanisms like biometrics. One efficient and low cost biometric modality is keystroke dynamics, which attempts to recognize users by their typing rhythm. It has been shown that the biometric features may undergo changes over time, which can reduce the predictive performance of the biometric system. Template update adapts the user model to deal with these changes and, therefore, decreases the predictive performance loss. Most of the studies in the literature only take into account samples classified as genuine to perform adaptation. This paper extends this common approach by proposing an original framework to make use of samples classified as impostors, too. This new approach, named Enhanced Template Update, uses all collected unlabeled samples to support the adaptation process. According to our experimental results, this new approach can improve the predictive performance when compared to current methods depending on the scenario. Some improvements on the visualization of results over time are also proposed during the analysis performed in this study. Although the proposed approach is evaluated on keystroke dynamics, it could also be applied to other biometric modalities.