Increase in level of detection of secret embedding into compressed JPEG-images based on selection of informative features

Abstract

The research of informational content of features in the spatial and the frequency domains at steganalysis of the compressed JPEG-images is presented. Modern steganalysis methods, based on which the feature set has been created, are considered. As a method of informative features selection has been applied the greedy algorithm with an exception which has allowed reducing number of features in the set from 43 to 26. The experiments have shown that steganalysis on the basis of informative features allows increasing the accuracy of correct classification of initial images to 94%, of images with attachments by the PM1 and the Jsteg methods by 25 and 18 percent respectively. The issue of steganalysis of images with artificially created distortions is mentioned. The experiments showed that steganalysis based on informative features allow classifying correctly such images with the accuracy 96%.