Automatic Siganture Verification using Baseline Comparison

OData support
Dr. Kővári Bence András
Department of Automation and Applied Informatics

Signature recognition is probably the oldest and yet the most common biometrical identification method, with a history of several hundred years. In contrast to other coexisting identification methods, like iris- or fingerprint-based authentication, signature-based verification does not require any special hardware, and the process takes considerably less time than a DNA-based verification.

Signatures are current and prevalent forms of identification, with a high legal acceptance, thus forged signatures not only cause economic problems – in the course of cheque frauds – but legal difficulties as well, hence it is vital to be able to differentiate between genuine and fake signatures. Over the decades two different types of computer based verification techniques arose: on-line and off-line signature verification.

The on-line methods take advantage of dynamic characteristics like velocity, acceleration or even the position and pressure of the pen, and provide reliable verification, with a high accuracy. However this information is not available in most real-life situations, as it requires particular capturing devices. In contrast the off-line methods do not require any special hardware, only the signature itself which makes them more user-friendly, but unfortunately more limited as well. In the past decade a bunch of solutions has been introduced to overcome the limitations of off-line signature verification and to compensate for the loss of accuracy, but only a few currently known methods can break the barrier of the error rate of 10 percent.

In this thesis we study the earlier signature verification techniques and their results, particularly the offline methods. After this we introduce a method for comparing shape-based features, and a feature – the baseline – which will help us improve the accuracy of an existing signature verification system.

After that we measure how precisely our approach can differentiate between genuine and forged signatures, we interpret those results, and finally, we suggest some future possibilities for improving the overall performance of our verification system.


Please sign in to download the files of this thesis.