Research On Blind Detection Based On Image Content

As the most important technical tools of image information security, Image steganalysis and digital forensics have become very attractive hotspots to researchers all over the world. Image steganalysis, as the opposite technology against steganography, aims at detecting, extracting, restoring and destroying the secret messages embedded into the cover images. Digital forensics is to identify the authenticity, integrity and origin of the digital images without embedding any information beforehand.Image steganalysis and digital forensics are carried out based on the analysis of the changes of the statistical features of the image data caused by secret message embedding or tampering. Image, carried the visual information by the space structure, is a local stationary Markov source. Consequently, it is significant to make a research on blind detection of image steganography and tampering based on the image content.This dissertation focuses on the study of blind detection of LSB matching embedding and tampering based on the image content. In this thesis, an image is assumed to be a local stationary Markov source, and some blind detection methods based on the study of statistical features of images and the mechanism of LSB matching embedding and tampering are proposed. The contributions obtained in this thesis can be summarized in the following aspects:1. The correlation between the complexity of the content and the statistical features as well as the texture of the image is analyzed based on statistical methods. The analytical results indicate that the statistical features of images with different content are distinct, and the complexity of the content could be measured by texture. The images could be divided based on the texture, and features can be extracted according to the statistical features of the local area.2. A high resolution real image database is set up, and the images in the database are classified according to the complexity of the content. The embedding existence features, including the center of mass of the histogram characteristic function and local extremum of histogram, are extracted, and the correlation between the embedding existence features and the statistical features is analyzed. The differences of the statistical features between the cover and stego images are compared. The analysis indicates that after embedding, the existence features are easier to be detected in the smooth area, which can be used for steganalysis.3. A LSB matching steganalysis method is proposed on the basis of the smoothness of the neighbourhood pixels. The smooth areas of the given image are chosen by the value of neighbourhood pixel differences. 15 features, including the smoothness of the image histogram and local co-occurrence matrix, pixel differences histogram as well as co-occurrence differences matrix, are extracted. The twice embedding process is used to eliminate the influence of different image contents, and the Fisher linear discriminator is applied for classifying. Experimental results show that the proposed method exhibits excellent performance on 3 databases.4. A method to detect the resampling manipulation based on texture complexity and singular value decomposition is proposed. The specific statistical changes brought into the linear dependencies of image pixels with different texture complexity by the interpolation process of resampling are analyzed by singular value decomposition. Features based on singular values are extracted, and support vector machine is applied for classifying. The performance shown in experimental results indicates the validity of the algorithm.Finally, the research work for this thesis is summarized and the further research topics and directions in the future of blind detection of information hiding and image tampering are discussed.