New Design of Markov Steganalytic Features

How to reduce Markov features dimensionality while keeping their steganalytic ability is an important issue in the field of steganalysis. This paper generalizes the statistical hypothesis that images are isotropic from spatial domain to Discrete Cosine Transform (DCT) domain, and provides a new design method. The proposed method is suitable for Markov features of different orders with respect to different sources of extraction and it can effectively deduce the dimensionality of Markov features. Concretely, it can reduce the dimensionality of traditional intrablock features by 36% and that of interblock features by 72%. Experimental results show that the proposed method can also enhance the features detection ability.