Steganalysis based on wavelet texture analysis and neural network

A new steganalysis technique is proposed on the basis of statistical analysis on the texture of an image for the detection of wavelet domain information hiding techniques. Wavelet coefficients in each sub-band of wavelet transform are modeled as a Generalized Gaussian distribution (GGD) with two parameters. It appears that those parameters are a good measure of image features and can be used to discriminate stego-images from innocent images. Neural network is adopted to train these parameters to get the inherent characteristic of innocent and stego-images. These experimental results have proved that this method is applicable for the detection of information hiding techniques in wavelet, and the proposed algorithm is more effective to previously existing techniques.