Investigating multi-touch gestures as a novel biometric modality

Abstract

We propose a new behavioral biometric modality based on multi-touch gestures. We define a canonical set of multi-touch gestures based on the movement characteristics of the palm and fingertips being used to perform the gesture. We developed an algorithm to generate and verify multi-touch gesture templates. We tested our techniques on a set of 22 different gestures. Employing a matching algorithm for a multi-touch verification system with a k-NN classifier we achieved 1.28% Equal Error Rate (EER). With score-based classifiers where only the first five samples of a genuine subject were considered as templates, we achieved 4.46 % EER. Further, with the combination of three commonly used gestures: pinch, zoom, and rotate, using all five fingers, 1.58% EER was achieved using a score-based classifier. These results are encouraging and point to the possibility of touch based biometric systems in real world applications like user verification and active authentication.