Research On Image Hiding Parallel Detection Technology

Information hiding and steganography detection technology is an important research direction of information security, it have developed rapidly in recent years. Because of digital image have a large amount of redundant information, so commonly used as steganography carrier. The corporate firewall and intrusion detection system have little effect on image steganography, so the steganalysis technology is particularly important for national security, army and so on. In this paper firstly briefly describes the development of information hiding and steganalysis technology, and then introduces the related technologies of image steganalysis, including the basic framework of image steganalysis and common detection features and classifiers. On those bases, we do some researches and improvements around feature extraction and classification methods of image steganalysis, the main work is as follows:(1)Detailed study and analysis of the PEV calibrated features and neighboring joint density features, pointed PEV calibration feature does not consider image steganography impact inter DCT block, and image steganography can cause discontinuities between DCT blocks, so the PEV calibrated features consider not comprehensive; Neighboring joint density features considered only the local statistical properties after image steganography, it lacks overall features after image steganography. In order to more fully describe the impact on the statistical characteristics of image steganography, we propose a new fuse characteristic. It includes both global features of PEV, such as histogram, variance etc. and also includes both features of intra DCT block and inter DCT block. In order to verifying the effectiveness of fuse characteristic, combined with SVM classifier to detect image steganography. Experimental results show that the fuse characteristic has increased the image steganography detection accuracy.(2)We firstly introduce sparse representation in the image steganalysis. For minimize l1 paradigm on solving sparse representation coefficient is lack of precision, we propose use vector total variation model as the objective function to solve sparse coefficient, and then we combined with PEV calibrated feature to detect image steganography. Experimental results have proven that this approach is better than l1 paradigm on the image steganography detection accuracy.(3)Thanks to the number of training set is large, lower image feature extraction efficiency and underutilized CPU multi-core and other issues, this paper proposes a feature extraction algorithm based on parallel OpenMP, then respectively using the serial and parallel algorithms for the three image sets extract image steganography detection feature. Experimental results show that the parallel approach to some extent reduces the feature extraction time, and improve the utilization of CPU resources.