Comparison of Efficiency of Statistical Models Used for Formation of Feature Vectors by JPEG Images Steganalysis
Nataliya R. Koshkina · Theoretical and Applied Cybersecurity · 2020
In order to build effective analytical systems for digital covers steganalysis in the given practical conditions, it is necessary to analyze and evaluate the quality of existing methods and components. Thus, it is necessary to compare the baseline characteristics of the available candidates in order to select the optimal components of the system. However, the usage of results from open scientific publications may lead to incorrect comparison due to differences in the conditions of numerical experiments. This study is based on the principle of checking the performance of statistical models for feature vectors formation under the same conditions. The case of JPEG images steganalysis with the usage of machine learning techniques is considered. The performance and detection accuracy of statistical models such as CHEN, CC-CHEN, LIU, CC-PEV, CC-C300, GFR, and DCTR in case of message hiding in the frequency domain of digital images are analyzed.The results of the study are numerical estimates of the performance and accuracy of SVM classification in binary and multilevel steganalysis modes.