Two Notes from Experimental Study on Image Steganalysis

摘要

In recent years, several advanced methods for image steganalysis were proposed. During research process, some concerns are more and more addressed by steganalyzer. In this paper, we focus on several of these concerns. The first one is how to utilize SVM classifier in practical steganalysis, we use clustering analysis to divide training samples and train several SVM for detecting stego image. In this part we also discussed building an image database that can be used for evaluating steganography/steganalysis fairly. The second is how to designed proper classifier for steganalysis, especially how to take information of cover/stego image pair into account. We will discuss several notions regard to these two concerns.