Use of keystroke dynamics and a keystroke-face fusion system in the real world

Abstract

In the field of keystroke biometrics, many systems have been developed using more traditional approaches and evaluated based on their accuracies. However, with the rise in popularity of neural networks, the research in the recent years focuses mainly on deep learning approaches. We test the accuracy and maintainability of both types of classifiers and try to test a new vector distance approach using a custom distance metric to see if there could still be room for improvement in the field of more traditional systems for keystroke-based authentication. We also test a keystroke dynamics and face recognition fusion classifier to see if a multi-modal system is perhaps the best compromise in terms of maintainability and performance. We find that there is still room for research on more traditional keystroke rhythm comparison techniques and conclude that by fusing those together with a face recognition pipeline we can achieve very good results in web authentication systems.