An Approach for Authenticating Smartphone Users Based on Histogram Features

In this study, we propose to adopt histogram features obtained from smartphone sensors for building authentication models, which could be used to nonintrusively authenticate smartphone users in varying operating scenarios (e.g. standing and sitting) when engaged in using stationary apps. We adopted two smartphone sensors, namely touchscreen and orientation sensor, to evaluate their feasibility. Consequently, sixteen touch-based features and thirty-three orientation-based features were separately used to construct two authentication models. To evaluate the performance of two constructed models, thirty-five subjects joined for collecting experimental data in two operating scenarios, standing and sitting. The experimental results showed that the equal error rate (EER) of touch-based model was approximately 6.56% with features extracted from ten flick touch gestures and reduced to approximately 3.05% with sixty flick touch gestures. For orientation-based model, the EERs were approximately 10.27% and 7.07%, separately. The results showed that the histogram features of the adopted two sensors are feasible for authentication purpose. Specially, this study further discusses the phenomenon of multiple behavioral pattern over the adopted two sensors caused among different operating scenarios, such as standing and sitting.