A blind image steganalysis based on features from three domains
Yuan Liu, Li Huang, Ping Wang, Guodong Wang · 2008
A new blind steganalyzer is constructed to markedly improve performance with higher universalness and detection accuracy. It merges Robert gradient energy in pixel domain, variance of Laplacian parameter in DCT(discrete cosine transform) domain and higher-order statistics extracted from wavelet coefficients as the feature vector of the proposed steganalysis algorithm, and BP(back propagation) neural network is applied as the classifier in this paper. Extensive experiments show the efficacy of our steganalyzer on a large collection of images and on three steganography algorithms. It can detect Jsteg, Stool with the accuracy of 90%.