Study On Steganalysis For Jpeg Images

Currently there are a lot of images on the Internet. Secret information can be hidden in these images, and cannot be noticed by the human eyes. Some criminals use the information hiding technique for secret information transmission. JPEG format image is a common image format on the Internet, which is widely used as carrier to achieve the purpose of transmitting secret information. Many steganography algorithms and steganography tools for JPEG images are proposed, such as Steghide, EMD. So the steganalysis technique for JPEG images is very significant.In this paper, steganography and steganalysis of JPEG images are mainly researched, and the main work of this paper is listed as follows:1) After information hiding using steganography algorithms or steganography tools, the quantized DCT coefficients histogram of JPEG image will be changed. Analysis the security of these steganography algorithms with these changes.2) Author analyzed the inadequacies of the steganalysis algorithm based on joint density, which is improved in this paper.28-dimensional joint density features are extracted, then Support vector machine (SVM) technique is adopted combing with the28-dimensional statistical features to accomplish effective detection for JPEG image. Experimental results indicate that the improved algorithm has a better performance. The feature sensitivity is analyzed, and the improved features are more sensitive especially for Steghide. Furthermore, the improved steganalysis algorithm for mixed training model has a better performance at low embedding rate, the detection accuracy is increased by about5%.3) Author analyzed the inadequacies of the steganalysis algorithm based on Markov, which is improved in this paper.28-dimensional Markov features are extracted, then SVM technique is adopted combing with the28-dimensional statistical features to accomplish effective detection for JPEG image. Experimental results indicate that the improved algorithm has a better performance. The improved features are more sensitive especially for Steghide. Furthermore, the improved steganalysis algorithm for mixed training model has a better performance at low embedding rate, the detection accuracy is increased by about7%.4) Author compared the sensitivity of improved joint density features and improved Markov feature, which can complement advantages and disadvantages each other. Author designed a steganalysis algorithm merging the two features, the detection accuracy will be further improved. Experimental results indicate that the merging feature has better performance. The merging features are more sensitive especially for Steghide and EMD. Furthermore, the merging steganalysis algorithm for mixed training model has a better performance at low embedding rate, the detection accuracy is increased by about9%.