A novel word-independent gesture-typing continuous authentication scheme for mobile devices

Abstract

In this study, we produce a new continuous authentication scheme for gesture-typing on mobile devices. Our scheme is the first scheme that authenticates gesture-typing interactions in a word-independent format. The scheme relies on groupings of features extracted from the word gesture after it has been reduced to parts common to all gestures. We show that movement sensors are also important in differentiating between users. We describe the feature extraction processes and analyse our proposed feature set. The unique process of our authentication scheme is presented and described. We collect our own gesture typing dataset including data collected during sitting, standing and walking activities for realism. We test our features against state-of-the-art touch-screen interaction features and compare feature extraction times on real mobile devices. Our scheme authenticates users with an equal error rate of 3.58% for a single word-gesture. The equal error rate is reduced to 0.81% when 3 word-gestures are used to authenticate.