Classification algorithms in online signature verification

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

Handwritten signature verification is the most commonly used and legally adopted biometric verification approach. Beside the traditional, paper based solution more and more adopters (including Hungarian banks, and multinational companies) use digital instruments to capture and verify their customers signatures.

During the past decades, several approaches were presented to verify online signatures, amongst them some were near to the graphologists with error rates around one-two percent. It is a hard task to reach that level of accuracy.

In my thesis work, I submit a solution based on the fusion of several approaches including the widely used Dynamic Time Warping, the more special Kolmogorov-Smirnov test and also count with the length of the signature in time.

After using the first forty signatures of the Signature Verification Competition 2004 to train my verifier. With it, I achieved error rates below three percent. These results are comparable to those of the leading verifiers in the field.

Downloads

Please sign in to download the files of this thesis.