Universal steganalysis scheme using support vector machines

Abstract

The purpose of a steganalysis scheme is to distinguish the stego-image from suspicious images. A well-designed steganalysis scheme must distinguish stego-media from cover-media with a better probability rather than random guessing. Most steganalysis schemes belong to the embedding specific approach. However, Farid initially proposed universal steganalysis schemes to detect messages hidden by various embedding algorithms. Because detection accuracy is related to the trained characteristic database, the selection of statistical features will be more important. In 2006, Lyu and Farid indicated that some statistical features are less important than others and that they do not affect the classified results. The proposed universal steganalysis scheme focuses on the differences of statistical features formed by embedding algorithms and applies a support vector machine to distinguish the stego-image from suspicious images. The proposed scheme is more practical than Lyu and Farid's schemes and experimental results show the performance of the proposed universal steganalysis scheme is also superior to those of the above-mentioned schemes.