Steganalysis of F5-like steganography based on selection of joint distribution features

Abstract

For steganalysis of F5-like steganography with two types of widely used joint distribution statistical features: co-occurrence matrix and Markov transition probability matrix, a feature selection and fusion method based on separability comparison is proposed in this paper. Firstly, the changing ways of F5-like steganography to image data is analyzed. Then, according to different affecting ways of the changing to two types of features, the separability of each feature component is analyzed and compared. At last, based on the comparison conclusion, a new feature is obtained by selection and fusion. Experimental results show that, compared with existing typical features, the proposed new feature can achieve higher steganalytic accuracy and enhance the detection reliability.