RECOGNITION OF DYNAMIC HANDWRITTEN SIGNATURES USING METHODS OF FUZZY SET THEORY

Abstract

We devoted this work to the recognition of handwritten signatures entered using a graphics tablet. We proposed a feature model of a handwritten signature that characterizes the dynamics of signature writing through various channels and uses methods of fuzzy set theory. We proposed an algorithm for creating a reference template for a handwritten signature that uses the potential method to construct feature membership functions and works even with a small training sample. We conducted studies in which we identified the optimal values of the degree of compactness used in constructing membership functions from a reference template, the form of membership functions of terms of linguistic variables used in constructing a feature model, and, in addition, obtained rational compositions of features that minimize recognition error. We proposed a modular structure of a software system for recognizing handwritten signatures entered using a graphics tablet. With the parameters found as a result of the research, we obtained an EER value of 0.36% and a type II error value of 0.2%.