Research On Detection Of Hidden Information In Spatial Domain Of Images

With the rapid development of network communication technology and the increasing popularity of digital multimedia, multimedia data can be transmitted in a quick and easy manner through the communication networks. However, at the same time, the increasingly important security issues are revealed. Especially, the fast development and wide application of information hiding technology lead to the transmission of confidential information becoming more covert. Military agencies, government departments, financial institutions and other key sectors involved in the people's livelihood take advantage of the data embedding technology to achieve the secure transfer of confidential documents. Meanwhile, this technique can easily be used by malicious individuals and groups for a variety of illegal purposes, thus posing serious threat to the national security, social stability as well as the personal privacy. Information detection technology attempts to break the information hiding technique, and hence, it is not only a mean of combating the illegal information transmission, but also an important measure to assess the security of the information hiding system. Thus, research on the detection of hidden information is significantly important to ensure the public safety, national security, social stability, and the normal economic development.As image contains much perceptually irrelevant or redundant information, it is well-suited for serving as "cover". As a result, the information hiding technique that takes an image as the cover has achieved the most prominent achievements, and become the most mature technique with most wide application. It is therefore necessary to conduct in-deep investigation into the image information detection technique. In this paper, we take the digital images as the research object, and the research includes information detection methods for spatial domain images, rate estimation of hidden information, and the feature selection and fusion methods that are suitable to the information detection. The main contributions of this paper are listed as follows.(1) The previous images detection methods usually suffer from the disadvantage that redundant features and the computational complexity increase with the increase of the dimension of the extracted features, and moreover, the high dimension of the features likely results in the dimensional disaster and affects the classification accuracy and other issues as well. In order to address this problem, we propose a feature selection method based on multicollinearity analysis, PCA (Principal Component Analysis) and Savage decision-making. Firstly, we analyze the multicollinearity among features to eliminate redundant features. Next, we implement the linear transform based on PCA and use the Savage decision-making to eliminate insignificant features. Finally, for the further reduce of features, we fuse the selected features, followed by selecting the principal features from the fused features to form a new feature set. The experimental results show that our method can reduce the redundant features, save the computational time, and make the detection more reliable.(2) By analyzing the impact of LSB replacing to the adjacent pixels of images statistical properties, a detection method based on statistical characteristics of adjacent pixels is proposed using the natural images statistic model. This method is theoretically proved to be effective. The experimental results show that this method is able to detect the random LSB replacing and sequential LSB replacing stegnography effectively. Meanwhile, by using the natural images statistic model, we introduce a detection scheme that is capable of accurately estimating the size of hidden secret message and also present the theoretical analysis.(3) In order to cope with the low detection result problem of LSB matching steganalysis for the uncompressed images, a new steganalytic method, which exploits the difference statistics of neighboring pixels, is proposed to detect the presence of spatial LSB matching stegnography. Stego images are generated through performing LSB matching operation to the test images. We take the changing rate of the pixels difference before and after the LSB matching stegnography together with the pixels difference before LSB matching stegnography as features, and the SVM is adopted to construct classifier. Experimental results demonstrate that the proposed method is efficient to detect the LSB matching steganography for both the compressed and uncompressed images. In addition, the experimental results have further demonstrated that the feature selection method proposed in chapter 3 is also effective to the detection of hidden information in the spatial domain.(4) Based on the problems that, for example, only a few steganalysis methods for palette images are available in current literature, the hidden information is hard to detect, and the existing detection techniques applied to the palette images directly is limited, a novel lifting integer Wavelet Transform based steganalysis algorithm is proposed to detect the hidden information in palette images. Firstly, we divide the image into non-overlapping 8x8 blocks and perform 5/3 IWT (Integer Wavelet Transform) based on lifting scheme to each block. Then, the average correlation is used to determine whether or not the secret message exists in palette images according to the correlation between the low frequency coefficients and high frequency coefficients. Experimental results show that the algorithm is able to better detect the hidden information in the palette images.