A Biometric Attack Case Based on Signature Synthesis

Abstract

One of the biggest challenges in on-line signature verification is the detection of malicious and skilled attacks. This paper proposes a new conceivable attack for an on-line signature biometric scheme. The attack consists in interpolating a smoothed 8-connected version of the forged signature and selecting the most relevant salient points, skipping those that belong to tremor or indecisive movements due to the faking procedure. The Sigma-Lognormal model is then used to synthetize the new on-line signature in the hope of obtaining an improved imitation. The experiments aim to prove that the False Acceptance Ratio (FAR) is significantly worsened with the Biosecure-SONOFF public online signature database. These results are expected to elicit new automatic signature verifiers able to cope with this new kind of attack.