Development of a steganalysis system for digital images based on a neural network classifier

The article discusses an approach to the implementation of a system for steganographic analysis of digital images based on a neural network classifier. It is used as a part of an integrated system for monitoring information security events of corporate infocommunication systems. As a basic structure for the neural network classifier, it is proposed to use a modified version of the convolutional neural network. Its preprocessing module implements the histogram method for analyzing the color and brightness characteristics of digital images. To automate the learning process of the neural network classifier, it is suggested to introduce a module for mass generation of stegocontainers with predefined values for the type and size of a digital image as well as for the size of the payload into the structure of the system being developed. Based on the developed structure of the steganalysis system for digital images, a factorial experiment was planned and conducted to evaluate the quality of the described neural network classifier in comparison with the known solutions of binary statistical classifiers. The choice of the area under the error curve (AUC ROC) as a metric for assessing the quality of classification is the main feature of the experiment. The results show that it is possible to use neural network classifiers to solve steganalysis problems, including their implementation in advanced information security tools.

1. Shipulin P. Steganografiya. Siberian Federal University; 2017. Available by: http://security.pmkb.sfu-kras.ru/blog/steganografiya/ (accessed on 20.02.2022). (In Russ.)

2. Genne O.V. Osnovnye polozheniya steganografii. Zashita informacii. Confident. 2000;(3):36–39. (In Russ.)

3. Kolesnikov A.A., Yandashevskaya E.A. Teoretiko-informatsionnyi podkhod k modelirovaniyu raspredelennoi steganograficheskoi sistemy s passivnym protivnikom. Sistemy upravleniya i informatsionnye tekhnologii. 2020;3(81):19–23. (In Russ.)

4. Bashmakov D.A. Metody i algoritmy vyyavleniya vstroennykh soobshchenii v prostranstvennoi oblasti nepodvizhnykh izobrazhenii pri maloi poleznoi nagruzke: dis. na soiskanie uchenoi stepeni kand. tekhn. nauk. Saint-Petersburg; 2018. 150 p. (In Russ.)

5. Grebennikov V.V. Steganografiya. Istoriya tainopisi. Moscow: LitRes: Samizdat; 2019. 160 p. (In Russ.)

6. Yandashevskaya E.A. Razrabotka podsistemy stegoanaliza tsifrovykh izobrazhenii na osnove svertochnoi neironnoi seti dlya obnaruzheniya i predovrashcheniya atak, ispol'zuyushchikh skrytye steganograficheskie kanaly. Doklady TUSUR (VAK). 2021;24(2):29–33. (In Russ.)

7. Fukunaga K. Introduction to statistical pattern recognition. Purdue university; 1972. 368 p.

8. Yandashevskaya E.A., Polunin A.A. Ispol'zovanie apparata svertochnykh neironnykh setei dlya stegoanaliza tsifrovykh izobrazhenii. Sbornik materialov Mezhdunarodnoi konferentsii «Ivannikovskie chteniya», Trudy ISP RAN. 2020;32(4):155–164. (In Russ.)

9. Kilbas I.A., Paringer R.A. Sravnenie tochnosti raspoznavaniya stsen i proizvoditel'nosti svertochnykh neironnykh setei. Nauki o dannykh: Sbornik trudov V Mezhdunarodnoi konferentsii i molodezhnoi shkoly «Informatsionnye tekhnologii i nanotekhnologii». 2019;740–747. (In Russ.)

10. Sikorskii O.S Obzor svertochnykh neironnykh setei dlya zadachi klassifikatsii izobrazhenii. Novye informatsionnye tekhnologii v avtomatizirovannykh sistemakh. 2017;(20):45–53. (In Russ.)