Evaluation of stability of swipe gesture authentication across usage scenarios of mobile device

Abstract

Background: user interaction with a mobile device predominantly consists of touch motions, otherwise known as swipe gestures, which are used as a behavioural biometric modality to verify the identity of a user. Literature reveals promising verification accuracy rates for swipe gesture authentication. Most of the existing studies have considered constrained environment in their experimental set-up. However, real-life usage of a mobile device consists of several unconstrained scenarios as well. Thus, our work aims to evaluate the stability of swipe gesture authentication across various usage scenarios of a mobile device.

Methods: the evaluations were performed using state-of-the-art touch-based classification algorithms--support vector machine (SVM), k-nearest neighbour (kNN) and naive Bayes--to evaluate the robustness of swipe gestures across device usage scenarios. To simulate real-life behaviour, multiple usage scenarios covering stationary and dynamic modes are considered for the analysis. Additionally, we focused on analysing the stability of verification accuracy for time-separated swipes by performing intra-session (acquired on the same day) and inter-session (swipes acquired a week later) comparisons. Finally, we assessed the consistency of individual features for horizontal and vertical swipes using a statistical method.

Results: performance evaluation results indicate impact of body movement and environment (indoor and outdoor) on the user verification accuracy. The results reveal that for a static user scenario, the average equal error rate is 1 and it rises significantly for the scenarios involving any body movement--caused either by user or the environment. The performance evaluation for time-separated swipes showed better verification accuracy rate for swipes acquired on the same day compared to swipes separated by a week. Finally, assessment on feature consistency reveal a set of consistent features such as maximum slope, standard deviation and mean velocity of second half of stroke for both horizontal and vertical swipes.

Conclusions: the performance evaluation of swipe-based authentication shows variation in verification accuracy across different device usage scenarios. The obtained results challenge the adoption of swipe-based authentication on mobile devices. We have suggested ways to further achieve stability through specific template selection strategies. Additionally, our evaluation has established that at least 6 swipes are needed in enrolment to achieve acceptable accuracy. Also, our results conclude that features such as maximum slope and standard deviation are the most consistent features across scenarios.

More information

Identifiers

ORCID for Richard Guest: orcid.org/0000-0001-7535-7336

Catalogue record

Export record

Altmetrics