Image-Based Blind Steganalysis Using Wavelet Statistics and Analysis of Variance

Applying wavelet decomposition to build high-scale statistical model for capturing statistical difference between cover images and stego-images. However, not all wavelet statistics are able to reflect well statistical change due to hidden message embedded. By exploring analysis of variance, the statistics that are more sensitive to hidden message are chosen as features of images. Kernel-based support vector machine is chosen as classifier to implement blind steganalysis of images. Experiment results show that our method can reach a high testing rate to hidden message of images.