Offline Signature Verification Using a Statistical Approach

Abstract

There is a growing interest in signature verification with the increasing number of transactions, especially financial, that are being authorized via signatures. Hence methods of automatic signature verification are essential if authenticity is to be verified regularly. In this research, two statistical approaches are used to develop an offline signature verification system. Data collection was done from 100 individuals. Everyone was asked to provide 12 samples of his/her original signature for training and testing processes. 600 forgeries were collected from three forgers and 6 forgeries were generated for each of the original signature samples. In this study features were extracted from the signatures after the preprocessing stage. Altogether 10 features were collected and those were used to verify the signatures. It was found that when there is a multicollinearity, Generalized Linear model by estimating parameters using generalized estimating equations is not appropriate to solve the above problem. Multicollinearity problem can be minimized using factor analysis and then generalized linear model was found to be a more effective approach. However, further research needs to be carried out to solve this problem.