Authentication method of information systems user by their handwriting with multi-stage correction of primary data

Abstract

The article is devoted to the biometric authentication of users, namely authentication by handwriting. In this paper, the relevance of creating a biometric authentication system of information systems users by their handwriting was argued. After that, there were determined a lot of characteristics of handwriting for their further use for authentication. Based on the analysis of the selected characteristics, there was determined their suitability for further use during user recognition. The method of authentication of information systems users by their handwriting and the method of the necessary primary processing of handwriting samples of information systems users were developed. Primary processing need is caused by the specific use, for the dynamic transfer of images to a computer, a graphic tablet (or other touch screen device). This processing is to remove erroneous data and correction of data to be used for recognition. There are five types of errors and three types of data correction in the work. To improve the recognition process, the image of the written key phrase, for further use, is divided into images of individual characters. Accordingly, the user is forced by the condition that the characters of the key phrase entered should be written separately from each other. During the recognition, the parameters of not all points of the image, but only the most significant, control points are analyzed. There are three types of control points in the work and the reasoned significance of using the most optimal algorithm for this. One of the types of neural networks, namely the probabilistic neural network, was chosen as the recognition mechanism. On the basis of the proposed methods, the software was developed, using which, first, a database of training samples handwriting of information systems was formed.  Then, a series of experiments was conducted to determine the effectiveness of the application of the developed methods and to identify the most significant, for proper recognition, user authentication system settings. In the end, it was concluded that, despite the fact that the methods proposed in this paper allow to achieve a sufficiently high probability of correct recognition of users of information systems, the search for more efficient recognition mechanisms and other parameters that significantly affect the probability of correct recognition remains very actual objective.