Study On Steganalysis Technology Based On JPEG Image

With the development of science and technology, we are in the digital age, many products are expressed based on the electronic information and digital technology makes the products become rich, which keeps their storage, replication and transmission becomes simple and fast. However, this has challenged the security of the digital products and caused a series of serious problems. In order to prevent the secret information eavesdropped and stolen, keeping the transmission safe, we can use information hiding technology. Steganography is an important branch of the information hiding technology, which plays an important role in the realization of secret information transmission, hiding message, and it has obtained widespread development. However, a lot of illegal users embed secret information into information and cover the correct secret information using of steganography technology, which endanger the national security and social stability. Therefore, steganalysis technology has attracted more and more attention.In this paper, we mainly analyze and improve steganography and steganalysis algorithm which are based on the JPEG image. Transmission of the image as the secret information carrier is very common, and JPEG image is one of the most common image formats, so JPEG image steganography and steganalysis technology has been paid much more attention on. The main work of this thesis and contribution are as follows:Firstly, after analyzing the exist steganography and steganalysis methods; we focus on the comparison of three steganalysis methods, which are based on two order statistics and SVM. They are respectively the inside block complexity using of three direction, the inter block complexity which using two directions, co-occurrence matrix method and Markov detection. Through the correct experiment result we can learn the rate of Markov detection is the highest. Secondly, based on Markov steganalysis algorithm, we proposed two improved methods, which are adding three moving directions and increasing the number of differential, in order to modify the feature vectors, makes the vector machine training accurately. The experiment results show that, for some kinds of steganography algorithm embedded images, such as F5, LSB and COX steganography algorithm, the improved Markov detection has higher accuracy and sensitivity.Thirdly, using Markov algorithm detect a various of images, such as fuzzy image, spliced image and median filtered image, to achieve classification detection and multi-class detection. After analyzing the research experiment results, we can see Markov and the improved Markov algorithm have higher accuracy.Fourthly, we extend the detection styles of Markov algorithm, such as image mixed feature vectors detection, which is the training set and test set images are embed secret information by LSB, COX and F5algorithm respectively, and using a variety of steganography algorithm repeatedly to embed secret information, chaos feature vectors are input to the SVM. We propose the content of our research is to further strengthen the accuracy of feature vectors, and to prepare for Markov detection improved.Finally, the paper is summarized and the prospective developing direction of the steganalysis technology is also put forward.