Neural network modelling of kinematic and dynamic features for signature verification

Additionally, some digitizers capture other function-based parameeters, such as the vertical pressure exerted by the pen tip, azimuthal and altitude angles of the pen, and even the pen's in-air trajectory. As a physiological biometric trait, a signature is used in various applications, including access control, commercial transactions, document forgery detection, and the provision of evidence in legal scenarios such as the verification of last wills [9]. In biometrics, where impostors may attempt to forge signatures with varying degrees of skill, robust verification methods are crucial. Since the execution of a signature inherently involves movements of the hand, arm, and forearm, it is hypothesized that these motions may contain kinematic and dynamic unique characteristic of the signer [7]. From a kinematic perspective, this action can be characterized by the arm's angular position, θ(t), and angular velocity, ω(t). Dynamically, these movements are facilitated by force torques, τ(t), applied at the joints. One method used to obtain this valuable biomechanical information involves a physical robot programmed to mimic the act of signing. While a robot's ability to accurately replicate these movements depends on its configuration, working area, and degrees of freedom, it can effectively capture kinematic and dynamic features during the process. However, accessing these robots is costly and cumbersome.