Research On Content Based Adaptive Image Steganography

Information hiding is one of the important research subjects on information security in the field of current information technology. Steganography is a main branch of information hiding and primarily applied to covert communication. It aims to transmit secret information unnoticeably through public channel by hiding secret information into the open digital medium. The converse analytic technology against steganography is called steganalysis, which detects the existence of hidden information, estimates its quantities and extracts it by analyzing the statistical features of cover and stego data. As the two opposite aspects of convert communication, steganography and steganalysis have been developing rapidly in antagonism.Large quantities of image data are stored or transmitted by Internet so image is the most frequently-used cover form of steganography. Steganography using image as cover is called image steganography. Image steganography actualizes covert communication by making use of the redundancy in human visual perception system and statistical redundancy of image data. Conventional steganography adopts fixed means to embed secretes in the whole cover image and the pseudo-random statistical features have a great effect on the correlation of cover image. So the steganalyst can detect the existence of hiding information by analyzing the difference between statistical features of cover and stego images. The visual sensitivity varies with the complexity difference of the observed objects and therefore many researchers proposed adaptive schemes based on human visual system, which have improved the capacity of covert communication system greatly under the condition of visual imperception. However, the developement of steganalysis based on statistical detection promote the security requirement of staganography to expand from"visual imperceptivity"to"statistical undetectability". The regional stationarity of image source allows staganographer adjust and control the parameter or the method of embedding automatically according to the content information of the specific cover image when secret information is being embedded in cover. By such a way the noise produced by embedding operation matches the cover image's content attributes, and therefore the effects on the image correlation caused by information hiding is decreased. This kind of steganography is called"content based adaptive steganography".A given cover image has determinacy signal features reflecting its content attributes. As it's difficult to depict the attributes of image content, for the need of academic study and application requirements of steganography, it's necessary to build a more reasonable technical model. Based on the variety and complexity of the cover date, and analysing the influence of embedding on the content attributes of the whole or different regions of a given image, the proposed new embedding methods should be with higher security in order to resist the steganalysis tools with higher efficiency and broader universal property. Therefore research on content based adaptive steganography has both academic and applied value.Based on the regional statistical stationarity of image, the influence of image's content attributes on the steganography's performance of resisting statistical detection is analyzed in this thesis. And based on the analyzing, the thesis focuses on building model of content based adaptive steganography, designing new adaptive algorithm and analyzing the security of adaptive steganography system. The main contributions of this thesis are summarized as follows:1. Using the image base established by ourselves as subject investigated, we analyze the statistical features of flat, artificial textured and natural textured image, discuss the influence of conventional staganography on statistical features of images, and study the relationships between the cover image's content attributes and digital staganography system's performance of resisting detection. The analytical results indicate that the existence trace is more obvious when embedding in strong correlation images while the quantities of embeded secrete information is the same. As for the same cover image, the influence on the statistical features of the flat regions will be bigger than that of the edge detail regions while the quantities of embeded secrete information is the same.2. The model of content based adaptive image steganography is discussed. We summarize that synchronization and transmission of side information are the two key points in building side information communication model of adaptive steganography, and propose to use wet paper code to solve the problems. Based on above discussion, side information communication model of adaptive steganography is built.3. Based on analyzing of complexity statistics of the cover image noise-like bit-plane block and pseudo-random data block, a new WPC(wet paper code) based bit-plane complexity segmentation steganography is proposed. Using the complexity statistics of bit-plane blocks as side information, the new scheme dynamically chooses the noise-like region of image bit-plane to embed secrete bits. Comparison experiments show that the proposed scheme is better than the classical and improved BPCS in aspect of visual imperceptibility and statistical undetectablity.4. By analyzing the security of HVS(human visual system)-based adaptive steganography, it is pointed out that "increasing embedding efficiency, controlling amplitude of modification and adaptive embedding mode" are three approaches to resist blind detection. Therefore new edge-based adaptive schemes in IWT(integer wavelet transform) domain and spacial domain is proposed. Using the statistics of IWT coefficients or detail component as side information, edges of image are chosen as regions to embed secrete information, at the same time, double-layered WPC is used to improve efficiency. Experiments show that either in IWT or special domain, the new algorithm behaves better in resisting blind steganalysis.5. Based on the Gaussian Mixture Model(GMM) of natural images, the security of adaptive steganography is analyzed. To the LSB(least significant bitp-lane) matching and SSIS(spread spectrum image steganography) method, it is concluded that the undetetablity of contend-based adaptive steganography, whose embedding parameter varies with the region of the cover image, is higher than non-adaptive schemes. And some instruction about how to design parameter of adaptive steganography and how to choose cover image is given in LSB matching mode.Finally, the research work for this thesis is summarized and the further research topics and directions in the future of adaptive steganography are pointed out.