Blind steganalysis algorithm based new calibration for JPEG images

To improve the embedding sensibility of calibrated feature in steganalysis, by studying the relationship between calibration technique and feature, a mathematical model for calibration based on the calibration classification is established, and a blind JPEG steganalysis algorithm based on the new calibration is presented. First we crop 4 pixels in the image and put forward a modified form of calibration, then the histogram characteristic is obtained according to the spatial distribution of the image features before and after calibration, and the Markov transfer probability matrix of the new calibration is calculated on the basis of redundancy. Finally, we fuse these features with the blocks feature and obtain the feature vector.Through the detection experiment of nsF5, Jsteg and MB1 algorithms with low embedding rates and among the feature vector, it is shown that this method has a better detection performance compared with those existing calibration methods. Its correctrate is more than 90％. The feature sets also show some complementary characteristics. It can be more stable and reliable in the different quality factors experiment.