Steganography In JPEG Images

Information security is the basic guarantee for national informatization process and the development of information industry. Multimedia content security is a vital part of information security. As an effective approach to achieve multimedia content security, steganography has been attracting more and more attention of the academic society and security departments. As JPEG image is the most popular image format transferring via internet, steganography in JPEG images bears broad application filed. Up until now, the approaches for improving security performance of steganography in JPEG images can be classified into two categories. Approaches in the first category engage in reducing the embedding distortion made on the cover image, while those in the second category engage in preserving the statistics of cover images. This thesis engages in these two categories, and the main contributions are in the following four folds.1. A steganographic scheme in JPEG images based on the genetic algorithm is pro-posed. Firstly, by carefully selecting coefficients for embedding and choosing the corresponding relationship between secret message bits and cover coefficients, this method can preserve the three histogram characteristics of JPEG images. Secondly, the genetic algorithm is employed to optimize the modification direction and em-bedding distortion is reduced. Experimental results show that our proposed method outperforms previous methods in preserving characteristics of both histogram and blockiness.2. The non-monotonic security performance of traditional YASS (Yet Another Se-cure Steganography) is mathematically analyzed, and a new method is proposed to further randomize the embedding blocks. Firstly, we analyze the relationship between the detection rate of YASS and its embedding rate, and that between the detection rate of YASS and the probability of coincidence of embedding block-s and JPEG blocks. Secondly, a further randomizing embedding blocks method is proposed, which overcomes the defects of high probability of coincidence of embedding blocks and JPEG blocks, and preserves the statistics of cover images. Experimental results show that this method can survive the specific steganalysis of YASS proposed by Li et al., and its security performance under JPEG domain universal steganalysis is monotonic. 3. A channel selection of YASS is proposed which minimizes embedding distortion. Distortion imposed to the final JPEG image is minimized by minimizing the distor-tion to the medium spatial image. This channel selection rule can not only improve the security performance of YASS, but also boost the performance of other im-provement of YASS. Experimental results show the effectiveness of this channel selection rule in both visual quality and security performance in resisting the four popular universal steganalysis.4. The relationship between the performance of the matrix encoding and its radix is studied, which is a commonly used and effective technique in reducing embedding distortion. Firstly, the term of change is redefined to reflect not only the number of modified coefficients but also the amplitude of modification. Secondly, based on mathematical analysis, it is proved that the embedding efficiency of ternary matrix encoding is higher than that of matrix encoding with radix larger than3. Moreover, experimental results show that the embedding efficiency of ternary matrix encoding is larger than that of binary matrix encoding. A conclusion is drawn that the em-bedding efficiency of ternary matrix encoding is higher than matrix encoding based on other radixes.