An application of linear mixed effects model to steganography detection

Current technology allows steganography applications to conceal any digital file inside of another digital file. Due to the large number of steganography tools available over the Internet, a particular threat exists when criminals use steganography to conceal their activities within digital images in cyber space. In this paper, a set of statistical features are generated using linear mixed effects models in conjunction with wavelet decomposition for image steganography detection. It is important to generate features capable of distinguishing between a set of clean and steganography images for steganalysts in commercial industry, Department of Defense, government as well as law enforcement. In the experimental results, seven sets of images are used to measure the performance of the proposed method, a clean set and two JPEG steganography methods with three different embedding file sizes to create steganography images. The number of correct predictions that an instance is clean or steganographic are improved by as much as 38% when using the proposed linear mixed effects models compared to the linear fixed effects models.