Keystroke Dynamics for Biometric Recognition in Handheld Devices

With the recent widespread usage of mobile devices there has been an increasing research for automated recognition systems which could be implemented with the limited resources offered by such tools. Specifically, several biometric recognition approaches have been proposed in literature, most of them relying on the cameras, microphones, or gyroscopes installed from factories in smartphones to acquire biometric traits such as face, speech, and gait. Another interesting modality which could be exploited in handheld devices, without requiring any additional hardware upgrade, is keystroke dynamics, with this term referring to the typing behaviour characterizing each user. This trait is especially suited to perform recognition on mobile devices, leveraging on data recorded while typing a password or a personal identification number (PIN) to access any physical or logical facility. This paper evaluates the feasibility of performing keystroke-dynamics-based biometric verification by modeling the acquired sequences of typed characters with Gaussian mixture models. Three different public datasets, comprising recordings related to both passwords and PINs, are exploited to testify the effectiveness of the proposed approach.