The Study Of Steganographic Codes And The Design Of Distortion Functions In Adaptive Steganography

The purpose of steganography is covert communication by embedding secret messages into public digital multimedia data. The current trend of development is adaptive steganography, which take the properties of cover fully into account and adaptively embed messages into the least suspicious region in order to improve the ability of resisting detection. Adaptive steganography include two key issues, namely steganographic code and defining distortion according to the properties of cover. The value of distortion denotes the embedding impact caused by changing the corresponding cover element. The larger value of distortion, the less suited for changing and vice versa. The purpose of steganographic coding is to embed messages as many as possible at the cost of distortion as less as possible, that is, how to improve the embedding efficiency. This dissertation focused on digital images and did some related researches aims to design secure adaptive steganographic methods. The main contributions of this dissertation are as follows:1. Propose an adaptive steganographic algorithm for spatial domain based on model and iteratively predicting texture. In order to guarantee the receiver to extract message correctly, we use the way, which select some seed pixels according to a shared key and predict remaining pixels’texture, to synchronize side information with receiver. Then, we execute predicting texture iteratively to locate the most complex regions as the regions into messages are embedded and compute the changing probability distribution of pixels based on their values of texture. Next, we adopt the Model-Based steganography framework and utilize the entropy decoder to accomplish the embedding and extracting. Experimental results show that the proposed algorithm can achieve high level of resisting detection.2. Construct one code used for wet-scale model, we named it Informed ZZW. Wet-scale model deals with the multilevel distortion in adaptive steganography coding. Informed ZZW translates multilevel distortion into two-level distortion which corresponds to the wet paper model. First, we construct multilayer wet channel by folding cover and compute their distortion by setting proper parameters. Then, embed messages using wet paper codes successively from outer channel until the inner channel. The "dry" elements and "wet" elements in each channel can been determined according to their changing dependency relation between wet channels. At last, only modifying the inner channel can satisfy all requests by other channels. The number of elements which need be modified in outer channel is just the total number. Experiments demonstrate that Informed ZZW can make full use of multilevel distortion to reduce the embedding impact on cover. Hence, Informed ZZW can be applied into adaptive steganography.3. Make a study on how to define distortion for spatial image and JPEG image. For spatial image, we find that the local complexity of textures is the primary factor that affects the distortion. First, we use a function to compute the complexity level of textures and divide the cover into two parts, complex region and flat region, by setting a proper threshold of complexity level. Then, a distortion function is presented based on the complexity level of textures. For JPEG image without the uncompressed raw image, we find that the absolute values of quantized DCT coefficients approximately obey power law distribution and the magnitude of DCT coefficients is the primary factor that affects the distortion. Heuristically by this, we present the distortion function based on the magnitude of DCT coefficients. At last, combining the distortion functions with syndrome trellis code, we propose the steganographic algorithms respectively. Experimental results demonstrate that the steganographic algorithms based on the presented distortion functions can achieve stronger security than existing typical steganographic algorithms.