Study On Image Steganalysis Algorithm Based On Image Recovery Technique

As the technology evolves,human life becomes more and more digitalized,and new information security problems emerges.Information hiding as a branch of information security technologies,hides confidential information into digital contents,such as images,audios,videos,and texts,to covertly transfer secret messages.Information hiding technology covers secret communications with strong security guarantee.It has been widely deployed in military and business domains.However,if information hiding technology is illegally used,it might threat the information security.Therefore,the research on effective detection methods for hidden information is very important.Images are an important carrier of human perceptible information and play critical role when humans communicate with each other.Images are the most common digital media in Internet.Because images have high information redundancy,and can be easily broadcast,they are suitable for information hiding.Nowadays,the research on information hiding in images leads to the most fruitful results and has beed widely used.Therefore,it is necessary to develop detection methods for image based information hiding.This thesis mainly focuses on steganography analysis in temporal images,using image recovery technologies.The main contributions are the follows:(1)Study and analyze the general math model used in steganography analysis,propose a framework for steganography analysis based on image recovery technology model.This framework considers the commonality between two models to derive steganography analysis feature data,and then build correlation model from these data,use feature dimension reduction technology to optimize feature subspace,apply machine learning algorithms to train steganography analysis model.(2)Based on the above steganography analysis framework,propose a steganography analysis algorithm.It first de-noise with WHMT algorithm,and then obtain the difference matrix by subtract the de-noised image from the image to be detected.Compute the co-occurrence matrix from the difference matrix,and then use the co-occurrence matrix as the steganography feature.Use K-W verification on steganography features for the first order evaluation to filter relatively outstanding features,and then obtain optimal feature subspace using PCA.Use SVM to train and test on subspace features.Experiments show that the algorithm proposed in this thesis has superior detection performance than other algorithms.