An application of Sparse Code Shrinkage to image steganalysis based on supervised learning
Michiharu Niimi, Hideki Noda · 2011
This paper proposes an image steganalysis based on supervised learning using Sparse Code Shrinkage as a feature of image data. Sparse coding represents source signal as the linear sum of basic images, and has the property that the coefficients of basic images are distributed as non-Gaussian. Sparse Code Shrinkage that is able to be regarded as a filter can effectively separate Gaussian distribution noise from sparse code coefficients. We assume that the degradation of image data by information hiding occurs as Gaussian noise. Therefore, the noise estimated by Sparse Code Shrinkage would be informative for image steganalysis. In the experiments, we show our method outperforms previous steganalysis methods for F5, StegHide, Spread spectrum image steganography.