Autmatized feature matching in siganture verification

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

It is an essential requirement that documents should be signed authentically. Verification of authenticity was easier in the case of hand-written documents. For a printed document the authenticity of the signer has to be verified by two witnesses. Currently the authenticity of the signer and the signature can be verified by a computer program.

I have chosen the feature matching problem as the topic of my thesis work. The meaning of feature matching is to find pairs to the chosen features and by that I could define the similarity of signatures. I investigated the assigned baselines and the in the signature appearing loops. I give the result of signature verification in a printed document for better checking. This document is also the part of my work.

I started to solve the problem by using the methods of the graph theory. The signature features are given, and I prepare a graph for both the baselines and the loops. For each graph edge I define the deviations of the endpoint properties, and from these I create a weighting value. For comparison I worked out two algorithms, which apply the weighting values and choose a feature for all investigated signatures.

I evaluated the results on several graphical test interfaces. Based on that the rate of success for baseline matching is 74%, and for loop matching it is 47%.

It is obvious from the above mentioned data that the error probability is lower in the case of matching baselines than loops. That means in case of loops further investigations are necessary.

Downloads

Please sign in to download the files of this thesis.