Deepkey: Keystroke Dynamics and CNN for Biometric Recognition on Mobile Devices

Abstract

The use of mobile devices is not anymore limited to personal communication but, especially with the widespread of smart devices, it enables a variety of operations like accessing the Internet, storing personal data, making payments, and gaining access to restricted services or areas. Being used for such sensitive applications, the implementation of secure mechanisms to access such devices is of paramount importance. As a matter of fact, most modern smartphones implement access control mechanisms based on either face, iris, or fingerprint biometric traits for secure user recognition, and several researches are conducted on further alternatives offering potential advantages in terms of convenience or performance. In this paper we study the effectiveness of employing keystroke dynamics to discriminate between genuine and impostor users of mobile devices. Specifically, deep learning techniques, based on convolutional neural networks, are here exploited to derive discriminative characteristics from the typing patterns of subjects entering personal identification numbers (PINs). A large database comprising keystroke dynamics acquisitions from 150 subjects, each typing two different kinds of PINs, either short or long, is exploited to conduct the performed experimental tests.