Generating off-line and on-line forgeries from on-line genuine signatures

Abstract

Improving the ability of forgery detection in Automatic Signature Verifiers requires databases with an extensive and realistic set of skilled forgeries. However, collecting skilled professional copies is not always possible in many studies, and most the publicly available databases are limited from that point of view. Recently, the Kinematic Theory of Rapid Movements has been used to improve the skillfulness from a given on-line skilled forgery. This paper proposes a method based on the Kinematic Theory of Rapid Movements to generate both on-line and off-line skilled synthetic forgeries from a single on-line genuine specimen. The method consists of three steps: 1. The genuine on-line signature is decomposed as a sum of overlapped lognormal strokes, 2. The trajectory is modified by distorting the virtual target points and the lognormal parameters, and 3. A new velocity profile is automatically synthetized from the 8-connected modified trajectory. Experiments with multiple on-line and off-line signature verifiers show that this method can generate skilled forgeries harder to detect.