RESEARCH OF ARTIFICIAL NEURAL NETWORK APPLICABILITY FOR USER`S ONLINE HANDWRITTEN SIGNATURE VERIFICATION
Read the full article
For citation: Dikiy D.I., Artemeva V.D. Research of artificial neural network applicability for user`s online handwritten signature verification. Scientific and Technical Journal of Information Technologies, Mechanics and Optics, 2017, vol. 17, no. 4, pp. 677–684 (in Russian). doi: 10.17586/2226-1494-2017-17-4-677-684
The paper considers features of artificial neural networks (ANN) application for online user`s verification by handwriting dynamics. This problem is urgent due to the development of cloud computing and machine learning, that even more often are used for pattern recognition. The review of the most widespread methods of digital user`s data pre-processing shows the relevance of ANN usage for verification. On the example of discrete Fourier transformation we made the experiments with different ANN structures and training algorithms. For these purposes we used the database of Signature Verification Competition (SVC) 2004. Cartesian coordinates and time parameters of trajectories have been taken from the database. The results of research show that artificial neural networks are able to solve that task, but at the same time with increasing the number of samples for study, success probability for legal users and malefactors also increases. For increasing of ANN performance we suggest the application of correlation analysis method for research data. It helps to increase the efficiency of artificial neural network. The false acceptance rate, or FAR, is about 2.26% and the false recognition rate, or FRR, is 12.6% with the use of correlation analysis method. At the same time, the task of distinguishing between a legal user and a malefactor, knowing how the signature or the password looks, is unsolved so far. In our opinion, the problem solution lies in the usage of additional parameters of handwriting dynamics (pressure, angle of inclination) for analysis.