A Novel Game-Theoretic Model for Content-Adaptive Image Steganography
Qi Li, Xin Liao, Guoyong Chen, Liping Ding · 2017
Content-adaptive image steganography means that steganographer chooses embedding positions based on image textures. Steganalyst can also focus on detecting these positions according to image textures. Game theory is preferred to analyze the above situation. However, in previous game models, steganalyst will mistakenly identify that no bit is embedded, when the secret bit is the same as the least significant bit of cover image. In this paper, a novel game-theoretic model based on secondary embedding is proposed to correct this judgment drawback. Both steganographer and steganalyst would change their choices to find new Nash equilibrium by using game theory. Co-occurrence matrix and point deviation degree are utilized for describing their choices. The occurrence number of each pixel pairs is calculated to constitute co-occurrence matrix, and then Euclidean distance between one point and adjacent points is computed to locate embedding positions. We finally draw a conclusion that in content-adaptive image steganography, steganographer should select embedding positions from both edge areas and smooth areas of digital images.