Surprising? Power of Local Features for Automated Signature Verification

Marcus Liwicki; Muhammad Imran Malik

In: Proceedings of the 15th International Graphonomics Society Conference. International Graphonomics Society Conference (IGS-11), June 12-15, Live Aqua Cancun, Mexico, Pages 18-21, International Graphonomics Society, 6/2011.


In this paper we report on the results of an automatic signature verification system on data of the ‘ICFHR 2010 4NSigComp’ forensic signature verification competition. The goal of this competition was to estimate the performance of automated systems in detecting skilled forgeries from genuine signatures of a reference writer. Unlike previous research in the field of signature verification, where the task was generally to separate the genuine signatures from the forged ones, another equally important category of forgery, namely the disguised signatures was also addressed in this competition. A disguised signature is a signature written by the authentic author but with the intention of possible denial at a later date. The system described in this paper did not participate in the competition, since it has been originally designed by the organizers of the competition. As an interesting outcome of the experiments, the system could achieve better equal error rates that any of the other submitted systems. The somewhat surprising fact is that the system has not been adapted to detect disguised signatures; it has originally been created for the task of detecting simulated and authentic signatures. We strongly believe that the main reason for the good performance is the difference that our system is relying on local features.


Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence