Image Steganalysis

With the development of multimedia technology, steganpgraphy is playing a more and more important role in information transfer process. Steganpgraphy is mainly embedding the information in the carrier, as far as possibly transmits the information through the public channels without causing the attention of the third party, it pays attention to the covert and security. However, some unscrupulous criminals improperly use of the technology to transmit the risk of the information. It results that country’s secrets, military intelligence and privacy are in trouble. As steganpgraphy’s opposite, Steganalysis’s purpose is to detect the media whether to hide the secret information and extract the secret information or destroy the information, to estimate the length of secret information or key.The digital image is often used as the original carrier for embedding information and transmission. A brief introduction to this dissertation is as follow:(1) Steganalysis algorithm based on a two layer difference model matrix is proposed. Steganalysis algorithm usually breaks the correlation between pixels. Considering the two order Markov transition probability is the analysis of the relationship between adjacent values, so we use of the two order Markov chain and put forward the model of the two layer difference matrix. Along the horizontal, vertical and diagonal directions, we extract the features from the two layer difference matrix. The features are merged to form the final feature sets. At last, we classify the original image and stego image using the SVM classification. Experiments show that this algorithm can effectively detect the original and stego images. At the same time, it also can detect the adaptive pixel pair matching steganpgraphy and typical steganpgraphy algorithm separately. And the detection result is better than that of SPAM.(2) A steganalysis algorithm based on extreme learning machine is proposed which is to classify a variety of steganography methods. And its structure was optimized to determine the node number in the hidden layer. The extreme learning machine is proposed based on the single hidden layer feedforward neural network. Setting the proper number of hidden layer nodes by optimizing the structure, it assigns the random values to the input vector and the hidden layer deviation. The final output weights are obtained by least square method. The whole process is finished in one time and does not need iterations. It has good generalization ability and learning performance. It not only improves the calculation speed but also has a good effect, especially used a wide variety of steganography classification method in hybrid. Experiments show that the classification can effectively classify the original and stego images. And the classification result is better than SVM.