Image universal steganalysis based on best wavelet packet decomposition

Based on the best wavelet packet decomposition of images, a new universal steganalysis method with high detection correct ratio is proposed. First, the best wavelet packet decomposition of image based on the Shannon entropy information cost function is made. Second, high order absolute characteristic function moments of histogram extracted from the coefficient subbands obtained by best wavelet packet decomposition are regarded as features. Finally, these features are processed and a back-propagation (BP) neural network is designed to classify original and stego images. Three different steganalysis algorithms for three different cases of background and application condition are presented. To validate the performance of the proposed method, a series of experiments are made for six kinds of typical steganography methods, i.e. LSB, LTSB, PMK, Jsteg, F5 and JPHide. Results show that, the average detection accuracy of the proposed method exceeds at least 6.4% and up to 15.4% and has a better universal performance than its closest competitors. Furthermore, the proposed method can provide reference for designing the pattern recognition and classification algorithm based on best wavelet packet decomposition.