Universal Steganalysis Using Multiwavelet Higher-Order Statistics and Support Vector Machines

Sanping Li, Yusen Zhang, Chun-hua Li, Feng Zhao

Abstract: In this paper, a new universal steganalysis algorithm based on multiwavelet higher-order statistics and Support Vector Machines(SVM) is proposed. We follow the philosophy introduced in Ref[7] in which the features are calculated from the stego image’s noise component in the wavelet domain. Instead of working in wavelet domain, we calculate the features in multiwavelet domain. We call this Multiwavelet Higher-Order Statistics (MHOS) feature. A nonlinear SVM classifier is then trained on a database of images to construct a universal steganalyzer. The comparison to the current state-of-the-art universal steganalyzers, which was performed on the same image databases under the same testing conditions, indicates that the proposed universal steganalysis offers improved performance.