SigVer3D: Accelerometer based verification of 3-D signatures on mobile devices

Abstract

We present SigVer3D -– a convenient authentication method for users of mobile devices with built-in accelerometers. The method works by analyzing streams of signals returned by a mobile device’s accelerometer when the user uses the device to draw his (her) signature in 3-D space. We cast authentication as a binary classification problem and train SVM classifiers to identify successful logins. We explore two types of features to represent signal streams, which can be computed very fast even in devices with limited processing power, and demonstrate their effectiveness using gesture data collected from a group of subjects. Experimental results show that the method can differentiate between genuine users and imposters with average EER (equal error rate) of 0.8%. Given the wide availability of accelerometers in mobile devices, the method provides a promising complement to existing mobile authentication systems.