The Role of Velocity in Online Signature Verification: Security Concerns

Abstract

In recent years, advances in machine and deep learning have significantly improved the performance of online signature verification. The present study investigates a security vulnerability in verifiers when only the velocity of the signatures is altered. We observed that the performance of verifiers is barely affected when the velocity profile of an online signature is forged, while its trajectory remains unaltered. Consequently, we hypothesize that the performance of online signature verifiers does not change when we modify velocity, one of the main features that describes human handwriting. To analyze this vulnerability, we modified the velocity profile of online signatures, making it uniform. We did this in three different ways: first, resampling the trajectory every Nth point, preserving the original signature duration. Second, resampling the trajectory at equidistant points, also preserving the original signature duration. Third, resampling all signatures in the database to have equidistant points, thereby normalizing their average velocity. We evaluated two verifiers, DTW and Deep Soft DTW with the DeepSignDB dataset, training with original velocity and testing with the forged velocity profile. The findings indicate that the performance of verifiers does not change significantly when the velocity of the signatures is manipulated. These findings underscore a security vulnerability in state-of-the-art online signature verifiers.