Nowadays, the biometric authentication is becoming more important. Especially the ones that work automatically and easily but also efficiently and safe at the same time. The signature-verification is the oldest and the most frequently used authentication method. It has not been really widely spread in the automated validation, in spite of the fact that it has more than 30 years of research experience. One of the reasons is that there is not any easy and reliable method yet, which does not require an expensive, specialized equipment.
In my research, I analyzed the current limitations of signature-verification using a public database. In addition, for that I have used the tools of Azure Machine Learning which made it possible to assess and compare a wide range of algorithms. My purpose was to exploit the resources and algorithm database which are the framework behind. Using these I can achieve more accurate results.
In my thesis I have examined the effectiveness of exiting algorithms to perform the classification problem. Furthermore I have analyzed the impact of different input data sets for the accuracy of the algorithm. In addition to the existing modules also introduce its own approach, and the results are compared to publicly available solutions.
Overall 20 signatories of 40 signing of 150 characteristics examining for the general methodology and algorithm choice conclusions can be drawn, which can greatly facilitate the development of more complete automated signature-verification systems.