Blind image steganalysis based on multi-domain feature scaling and BP network

To improve the correct detection ratio and expand the application scope of universal blind detection for image steganography,a blind steganalysis method based on image multi-domain features scaling is presented.This method extracts features from spatial,DCT(discrete cosine transform) and DWT(discrete wavelet transform) domain respectively,including a special statistical irrelevance of adjoin pixels pairs,the variance of scale factor of Laplacian distribution of DCT coefficient macro-block,and higher-order statistics of DWT coefficient.Scaling these features according to their domain and combining them,a 26-D feature vector for each image can be obtained,and then a back-propagation(BP) neural network can be designed.A series of experiments validate the performance of the proposed method for eight kinds of typical steganography of BMP and JPEG images.Results show that this method can make a reliable blind detection for various kinds of typical steganography.