MobileTouchDB: Mobile Touch Character Database in the Wild and Biometric Benchmark

Abstract

In this paper, we introduce a new database of mobile touch on-line data named MobileTouchDB. The database contains more than 64K on-line character samples performed by 217 users, using 94 different smartphone models, with an average of 314 samples per user. In each acquisition session, users had to draw all numbers (from 0 to 9), upper-and lower-case letters (54), different symbols (8), and passwords composed of 4 numbers (6). Regarding the acquisition protocol, MobileTouchDB comprises a maximum of 6 captured sessions per subject with a time gap between them of at least 2 days. This database studies an unsupervised mobile scenario with no restrictions in terms of position, posture, and devices. Users downloaded and used the acquisition app on their own devices freely. In addition, we also report a benchmark evaluation of biometric authentication on MobileTouchDB, providing an easily reproducible framework for two different scenarios of biometric user authentication: i) based on one character, and ii) based on character combinations. The database was collected with three main goals in mind: i) analyse the discriminative power of novel human touch interaction dynamics, ii) enhance traditional password authentication systems through the incorporation of touch biometric information as a second level of user authentication, and iii) analyse the way we interact with mobile devices on a daily basis in order to enhance continuous authentication systems. MobileTouchDB is publicly available in GitHub.