Image-based Sparse Steganography Algorithm Research

With the rapid development and application of Internet, the issue of information security has become visible. As a main method of hidden communication and protection of knowledge copyright, technology of information hiding has been widely used and developed.Sparse representation of images over redundant dictionaries is a rapidly evolving field, with promising results in many applications including compression, image denoising, feature extraction and image retrieval. In this thesis, we focus on the research of sparse representation and its applications in steganography. Based on the existing research achievements in sparse representation, we proposed a new steganography based on image sparse representation, fully considering the importance of redundant dictionary and decomposition coefficients. Some works have been completed as follows.(1) A steganography algorithm based on image sparse representationBase on the theory of sparse representation, we represent a new steganography based on image sparse representation. Using binary redundant dictionary and matching pursuit method with non-zero constraint, the algorithm successfully overcomes the instability problem caused by insertion of secret data, and guarantees correct extraction of secret data. Any coding techniques in pixel domain can also be applied to our algorithm. Experimental results show that steganography in sparse domain can achieve better anti-detection ability than that in pixel domain.(2) Research on redundant dictionary in steganography based on sparse representationRedundant dictionary is an important issue in steganography, and there is little research on it. We systematically discuss and compare different redundant dictionaries'impact on performance of the steganography based on sparse representation for the first time. Here, the redundant dictionaries are built by the methods based on mathematic model, combining several basic dictionaries and machining learning, respectively. The design rules and requirements for three kinds of dictionaries, which can make the steganographic algorithm perform better, are concluded. We can see from the experiments, when employing the dictionary designed by learning method, the algorithm can achieve higher embedding capacity and better security ability.(3) Research on decomposition coefficients in steganography based on sparse representationDecomposition coefficients are another important problem, and they have great relationship with image statistic feature, having different sensitivity to steganalysis. We propose an embedding rule in sparse domain, which fully considering the importance of image decomposition coefficients. The embedding rule selects important coefficients for hiding secret data in order to improve the security ability of the steganographic algorithm. The experimental results show that, the data hiding algorithm using our embedding rule has stronger anti-detection ability.