Article 5217

Title of the article



Volchikhin Vladimir Ivanovich, Doctor of engineering sciences, professor, President of Penza State University (40 Krasnaya street, Penza, Russia),
Ivanov Aleksandr Ivanovich, Doctor of engineering sciences, associate professor, head of the laboratory of biometric and neural network technologies, Penza Research Institute of Electrical Engineering (9 Sovetskaya street, Penza, Russia),
Vjatchanin Sergej Evgenyevich, Associate professor, head of sub-department of radio and satellite communications, Faculty of Military Education, Penza State University (40 Krasnaya street, Penza, Russia),
Malygina Elena Aleksandrovna, Candidate of engineering sciences, research worker, the interindustrial laboratory of biometric device testing and technology, Penza State University (40 Krasnaya street, Penza, Russia),

Index UDK





Background. The aim of the work is to create an artificial neuron, which is analogous  to the "Cramer – von Mises" statistical criterion, for synthesizing neural network biometrics-code converters from these neurons.
Materials and methods. Previously the authors proved that the "Cramer – von Mises" criterion on small test samples works better than the Chi-square statistical test. It means that instead of neural network radial basis functions there can be used artificial neurons analogous to the "Cramer – von Mises" criterion. The inputs of the new type of neurons do not receive ordinary biometric parameters, but probabilities of their occurrence in a learning sample.
Results. From quadratic forms the synthesized artificial neuron inherited the learning algorithm’s stability that eliminates the problem of training large networks of neurons. As well as all quadratic neural network functionals, the "Cramer – von Mises" neuron networks give the output code with low entropy for “Foe” images due to the lack of balance of "0" and "1" states in its bits. This shortcoming is suggested to be eliminated by applying an output quantizer at the output of the "Cramer – von Mises" neuron adders with three output states.
Conclusions. The introduction of a quantizer with three stable states into the composition of the "Cramer – von Mises" probabilistic neuron allows to reach the highest level of entropy for examples of “Foe” images. This effect occurs because of the doubling of output bits of the output code and their almost total balance as of bits "0" and "1". As a result, the "Cramer – von Mises" neuron network turn out to be much more effective than other networks of other known quadratic functionals. Moreover, their output entropy code is higher than that of neural networks, formed and trained according to the state standard GOST 52633.5. This allows to examine the "Cramer – von Mises" neuron networks as a prospect for the next generation of neural network biometrics-code converters. 

Key words

neural network biometrics-code converter, biometrics, "Cramer – von Mises" statistical criterion

Download PDF

1. R 50.1.037–2002. Rekomendatsii po standartizatsii. Prikladnaya statistika. Pravila proverki soglasiya opytnogo raspredeleniya s teoreticheskim. Ch. I. Kriterii tipa χ2. [Standardization recommendations. Applied statistcs. Validation rules for experimental and theoretical matching. Part I. χ2 criteria]. Moscow: Gosstandart Rossii, 2001, 140 p.
2. R 50.1.037–2002. Prikladnaya statistika. Pravila proverki soglasiya opytnogo raspredeleniya s teoreticheskim. Ch. II. Neparametricheskie kriterii [Applied statistcs. Validation rules for experimental and theoretical matching. Part II. Nonparametric criteria]. Moscow: Gosstandart Rossii, 2002, 123 p.
3. GOST R 52633.5–2011. Zashchita informatsii. Tekhnika zashchity informatsii. Avtomaticheskoe obuchenie neyrosetevykh preobrazovateley biometriya-kod dostupa [State standard GOST R 52633.5–2011. Information protection. Information protection technique. Automatic learning of neural network biometrics-access code converters]. Moscow, 2011.
4. Khaykin S. Neyronnye seti: polnyy kurs [Neural networks: a complete course]. Moscow: Vil'yams, 2006, p. 1104.
5. Galushkin A. I., Tsypkin Ya. Z. Neyronnye seti: istoriya razvitiya [Neural networks: the history of development]. Moscow: Radiotekhnika, 2001, 840 p.
6. Kulagin V. P., Ivanov A. I., Vyatchanin S. E. Problemy avtomatizatsii i upravleniya v tekhnicheskikh sistemakh: sb. st. XXXII Mezhdunar. nauch.-tekhn. konf. (g. Penza, Penza 6–8 iyunya 2017 g.) [Automation and control problems in technical systems: proceedings of XXXII International scientific and technical conference (Penza, 6th-8th June 2017)]. Penza: Izd-vo PGU, 2017, vol. 1, pp. 116–119.
7. Gil'mutdinov A. Kh., Kachaykin E. I., Ivanov A. I., Bezyaev A. V. Nelineynyy mir [Nonlinear world]. 2017, vol. 15, no. 3, pp. 3–10.
8. Volchikhin V. I., Ivanov A. I., Pashchenko D. V., Akhmetov B. B., Vyatchanin S. E. Izvestiya vysshikh uchebnykh zavedeniy. Povolzhskiy region. Tekhnicheskie nauki [University proceedings. Volga region. Engineering sciences]. 2017, no. 1 (41), pp. 5–15.


Дата создания: 07.11.2017 10:45
Дата обновления: 07.11.2017 14:26