A steganalysis algorithm integrating resampled image multi-classification

Abstract

When steganalysis performed on heterogeneous images made up by different resampled images and raw single-sampled images, the difference of statistical properties between which can caused “mismatch” between training and testing images in steganalytic classifier. Therefore, the detection performance of the classifier decreases. The problem above limits the application of the existing steganalysis algorithms in practical networks. In this study, a multi-classifier based on SVM is constructed to perform multi-classification on the resampled image, and a steganalysis algorithm integrating resampled image multi-classification is proposed. The algorithm prevents the "mismatch" between the training image and the testing image, and improves the detection performance of steganalysis algorithm under the condition of hybrid heterogeneous images. Finally, the effectiveness of the algorithm is proved by experiments.