Quality indices of dynamic signature recognition systems based on naive Bayes classifier and neural network

Methods and systems of information protection, information security


Gurakov M. A.*, Krivonosov E. O.**, Kostyuchenko E. Y.***

Tomsk State University of Control Systems and Radioelectronics, 40, Lenin str., Tomsk, 634050, Russia

*e-mail: g.mishell@gmail.com
**e-mail: egor-yrga@mail.ru
***e-mail: key@keva.tusur.ru


The research aims at creation of an authentication system based on the traditional password protection with biometric addition, providing security indicator improvement compared to the original version. This addition is a dynamic handwritten digital signature authentication based on neural networks and Bayesian classifier combining. The paper considers the possibility of combining classifier based on neural networks with Bayesian classifier by linear combination of outputs to reduce the overall probability of error classification regardless the error types. Within linear combination, neural network plays the leading role, since the results of neural network match those of linear combination. Results similarity was confirmed by comparison with expected values. It means that a linear combination without using errors of types 1 and 2 taken into account separately does not make sense because of the possibility of implementing neural network instead of linear combination. Thresholds of authentication for using errors of 1-st and 2-nd types taken into account separately accounting were created. The systems based on modified naive Bayes classifier and neural network, depending on the thresholds of authentication, was implemented. Software module for calculating errors of the 1-st and 2-nd type in neural network and Bayesian classifier was developed. The behavior of user dynamic parameters within signature recognition systems was presented. Performance evaluation of authentication system in the form of probabilities of errors of 1-st and 2-nd types was obtained.


identification, signature, Bayes classifier, neural network, integration, linear dependence, type I and type II errors


  1. Hodashinskiy I.A., Savchuk M.V., Gorbunov I.V., Meshcheryakov R.V. Doklady Tomskogo gosudarstvennogo universiteta sistem upravleniya i radioelektroniki, 2011, no. 2(24), pp. 236-248.

  2. Gurakov M.A., Krivonosov E.O. Materialy konferentsii uchastnikov gruppovogo proyektnogo obucheniya TUSUR, 2014, URL: https://storage.tusur.ru/files/10909/KIBEVS-1005_Autentifikatsia_polzovatelya_po_dina.pdf (accessed 1.06.2015).

  3. Subbotin S.V., Bol'shakov D.Y. Zhurnal Radioelektroniki, 2006, no. 4: http://jre.cplire.ru/jre/oct06/2/text.html.

  4. McCallum, A. and Nigam K. “A Comparison of Event Models for Naive Bayes Text Classification”. In AAAI/ICML-98 Workshop on Learning for Text Categorization, pp. 41-48. Technical Report WS-98-05. AAAI Press. 1998.

  5. Gmurman V.E. Teoriya veroyatnostej i matematicheskaya statistika (Theory of Probability and Mathematical Statistics), Moscow, Vysshaya shkola, 2003, 479 p.

  6. Doroshenko T.Y., Kostyuchenko E.Y. Doklady Tomskogo gosudarstvennogo universiteta sistem upravleniya i radioelektroniki, 2014, no. 2(32), pp. 219-223.

  7. Kostyuchenko E.Y. Identifikatsiya nepreryvnykh biometricheskikh signalov na osnove neyronnykh setey (Identification of continuous biometric signals based on neural networks), Doctor’s thesis, Tomsk, 2010, 171 p.

  8. Kostyuchenko E.Y., Meshcheryakov R.V., Kraynov A.Y. Doklady Tomskogo gosudarstvennogo universiteta sistem upravleniya i radioelektroniki, 2010, no. 1(21), pp. 118-220.


mai.ru — informational site MAI

Copyright © 2000-2021 by MAI