Keystroke biometrics, as an authentication method with advantages of no extra hardware cost, easy-to-integrate and high-security, has attracted much attention in user authentication. However, a mass of researches on keystroke biometrics have focused on the fixed-text analysis, while only a few took free-text analysis into consideration. And in the field of free-text analysis, most researchers usually devote their efforts to extracting the most appropriate keystroke features on their own experience. These methods were inevitably questionable due to their strong subjectivity. In this paper we proposed a multi-user keystroke authentication scheme based on CNN model, which can automatically figure out the appropriate features for the model, adjust and optimize the model constantly to further enhance the performance of model. In the experiment on a small sample set, the performance is improved more than 10% compared with the benchmark. Our model achieves an average recognition accuracy of 92.58%, with FAR of 0.24% and FRR of 7.34%.
Lv, Kun; Liu, Jiafen; Tang, Ping; and Li, Qing, "Keystroke Biometrics for Freely Typed Text Based on CNN model" (2018). PACIS 2018 Proceedings. 291.