Robotic arm motion for verifying signatures

Abstract

This paper proposes a novel set of function-based features for dynamic signature verification. They are inspired on the human stance and the variations in the angles of the arm joints when signing. Specifically, we propose to convert the trajectory and pen-tip altitude and azimuth from an online signature into the required sequence of an anthropomorphic robot poses to reproduce such signature. Then, these new robotic sequences are evaluated in an on-line automatic signature verifier. The robotic arm poses are defined by the angles of its joints. These values are worked out by means of the homogeneous transformation matrices between the different coordinate frames of each joint by using the Denavit-Hartenberg (DH) parameterization. The sequences of joint angles represent the new function-based feature space that we propose for signature verification. Our experimental results on the MCYT-100 corpus highlight the effectiveness of robotic-based features in signature verification. Moreover, competitive performances are achieved in random and skilled forgery experiments, compared to previous approaches.