Signatures are widely used for identification, therefore the investigation of their originality is critical. The automation of the signature verification process is a very important research field, because it can significantly speed up the verification process.
The automation of the signature verification process can be done online or offline. In case of online, we are using special tools for recording the signatures, therefore we can gain detailed information about them. In the offline case, we can only verify the signature based on its image, therefore there is no need for any special tools then. A modular system is developed at the Budapest University of Technology and Economics Automation and Applied Informatics Department, which performs the offline verification process. One part of this modular system is the matching process, when the features, which were extracted from the signatures' images, are matched with each other.
Matching the features to each other can be really problematic, since it is not sure, that we will find a pair for every features in every signature. It makes it also harder, that the feature extraction is processed automatically, which is imprecise and often detect significantly diverse features. To lower the impact of these problems, I always tried to find the optimal matching for a set of signatures, instead of for only one pair of signatures. This problem can originate in a well-known mathematical problem, the n-dimension matching problem, or more generally in the Set packing problem, that are also introduced in my thesis. Since the investigated problem is NP-hard, I am using approximation algorithm to determine the best matching. Lastly, I defined a metrics for evaluating the preciseness of the matchings and I also provide a feature that makes it possible to export the results into an Excel table.
Right now I am working on the final integration into the system of my matching algorithm, that would sensibly improve the preciseness of the given verification system.