Online signature verification using the TensorFlow framework

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

Machine learning solutions became one of the most popular topics of informatics lately. There are numerous problems, which these kinds of solutions could be very useful for, because men or scripted algorithms of other kind could not solve, or simply would need much more time.

Signature verification can also be solved with using machine learning solutions. Verification with signature belongs to the less secure biometric verification methods, but on the other hand it is pretty easy to implement. I am going to discuss the topic of online signature verification in my thesis.

During my work, I am going to use DTW, which is an algorithm for calculating distances between time series. I am going to compare three classifiers, the first one being a classifier implemented trying to reproduce a contest winning paper. It uses a linear user independent classifier with PCA. The other two implement one classifier for each user using only DTW distances and a threshold. From these three the one reproducing the paper seemed to be bringing the best results, being short only minimally on the result of the paper.

My thesis could give an introduction into the topic of machine learning and online signature verification.

Downloads

Please sign in to download the files of this thesis.