View article

In the paper, we advocate a new approach to blind steganalysis based on classifying higher-order statistical features derived from an estimation of the stego signal in the wavelet domain. The proposed approach is flexible and enables reliable detection of presence of stego messages embedded using a wide range of steganographic methods that include±1 embedding (LSB matching), LSB embedding, Stochastic Modulation, and others. The method is tested on raw, never compressed digital camera images, scans of photographs and films, as well as preprocessed (JPEG compressed, downsampled) images. The performance of this method is compared to the current state-of-the-art steganalytic methods for additive steganography.