Usage of quadratic form networks for users' recognition by dynamic biometrie images

Static biometrie images are not private and may be copied to make a physical and digital stimulant without an owner being aware of it, therefore the search for efficient solutions of personal authentication via dynamic biometric characteristics is still in process. A series of computational experiments based on biometric data obtained through a handwritten signature and keystroke dynamics is carried out. In the experiment the perceptron, quadratic form networks and Chi-module functionals were used. The paper proposes to adapt the algorithm of training the perceptron according to the Russian State Standard GOST R 52633.5-2011 to set quadratic form networks. A number of errors while verifying a person by handwriting dynamics is 1%, and while authenticating a person in a two-factor mode (a handwritten signature and keystroke dynamics) is 0.31%.