SEGAUTH: A Segment-based Approach to Behavioral Biometric Authentication

Many studies have been conducted to apply behavioral biometric authentication on/with mobile devices and they have shown promising results. However, the concern about the verification accuracy of behavioral biometrics is still common given the dynamic nature of behavioral biometrics. In this paper, we address the accuracy concern from a new perspective-behavior segments, that is, segments of a gesture instead of the whole gesture as the basic building block for behavioral biometric authentication. With this unique perspective, we propose a new behavioral biometric authentication method called SEGAUTH, which can be applied to various gesture or motion based authentication scenarios. SEGAUTH can achieve high accuracy by focusing on each user's distinctive gesture segments that frequently appear across his or her gestures. In SEGAUTH, a time series derived from a gesture/motion is first partitioned into segments and then transformed into a set of string tokens in which the tokens representing distinctive, repetitive segments are associated with higher genuine probabilities than those tokens that are common across users. An overall genuine score calculated from all the tokens derived from a gesture is used to determine the user's authenticity. We have assessed the effectiveness of SEGAUTH using 4 different datasets. Our experimental results demonstrate that SEGAUTH can achieve higher accuracy consistently than existing popular methods on the evaluation datasets.