A wavelet-based blind JPEG image steganalysis using co-occurrence matrix

A new approach to blind detection of JPEG images having hiding information, which is based on the feature of wavelet domain is put forward. The specific way of detection is as follows: First, decompose an image by using wavelet, calculate the co-occurrence matrix of the adjacent wavelet coefficients, and apply Laplace transform to the co-occurrence matrix. Then take the Laplace-transformed variances and characteristic function (CF) moments of the co-occurrence matrix as the statistical feature, and choose BP neural network classifiers to classify and detect. Experiments of detecting the four typical steganographic algorithms of Jsteg, F5, Jphide and Outguess in such JPEG images have been carried out at different embedded ratios by using the method mentioned in this article, and the results show that the blind detecting method has the higher accuracy and it has higher calculating speed compared with the typical blind detecting methods.