Studying the generalized coordinates and torques in on-line signature verification

Abstract

Biometric verification through handwritten signatures is a widely accepted behavioral approach in security systems. Traditionally, on-line signature verification has relied on kinematic data captured in Cartesian space. In this work, instead of using conventional spatiotemporal sequences, we employ the joint angle kinematics of a simple two-degree-of-freedom robotic model that performs the signatures. Additionally, we incorporate dynamic information by estimating the torques of the robotic arm using the Lagrangian formulation. We analyze the effectiveness of using generalized coordinates and torques as new feature spaces for on-line signature verification. Our experiments across several public databases demonstrate that combining these robotic features provides improvements over traditional timing-based features.