Steganalysis algorithm based on Cellular Automata Transform and Neural Network

In this paper, a new steganalysis method based on Cellular Automata Transform (CAT) is presented. CAT is used for feature extraction from stego and clean images. For that purpose, three levels CAT is applied on images and 12 sub-bands are generated for feature extraction. With adding the original image, 13 sub-bands are be used in feature extraction phase. In the next step, three moments of characteristic function (CF) are used as feature vector for every image (stego or clean image). At the end, Neural Network (NN) is applied as classifier. This supervised learning method uses these features for classifying the input image into either stego-image or clean-image. The performance of this algorithm is verified using some test samples. The results of our empirical tests show that detection accuracy of our method reaches to 93% for breaking MB2 and 91% for breaking LSB. Therefore the proposed method is a blind steganalysis method that can be used for broking some steganography methods.