Exploring a statistical method for touchscreen swipe biometrics

The great popularity of smartphones and the increase in their use in everyday applications has led to sensitive information being carried in them, such as our bank account details, passwords or emails. Motivated by the limited security of traditional systems (e.g. PIN codes, secret patterns), that can be easily broken, this work focuses on the analysis of users normal interaction with touchscreens as a means for active authentication. Given the frequency in which touch operations are performed, characteristic habits, like the strength, rhythm or angle used result in discriminative patterns that can be exploited to authenticate users. In the present work, we explore a statistical approach based on adapted Gaussian Mixture Models. The performance across different kinds of touch operations, reveals that some gestures hold more user-specific information and are more discriminative than others (in particular, horizontal swipes appear to be more discriminative than vertical ones). The experimental results show that touch biometrics have enough discriminability for person recognition and that they are a promising method for active authentication.