Where Usability and Security Go Hand-in-Hand

Abstract

Gestures have recently gained interest as a secure and usable authentication method for mobile devices. Gesture authentication relies on recognition, wherein raw data is collected from user input and preprocessed into a more manageable form before applying recognition algorithms. Preprocessing is done to improve recognition accuracy, but little work has been done in justifying its effects on authentication. We examined the effects of three variables: location, rotation, and scale, on authentication accuracy. We found that an authentication-optimal combination (location invariant, scale variant, and rotation variant) can reduce the error rate by 45.3% on average compared to the recognition-optimal combination (all invariant). We analyzed 13 gesture recognizers and evaluated them with three criteria: authentication accuracy, and resistance against both brute-force and imitation attacks. Our novel multi-expert method (Garda) achieved the lowest error rate (0.015) in authentication accuracy, the lowest error rate (0.040) under imitation attacks, and resisted all brute-force attacks.