Research On The Key Technologies And Methods Of Digital Image Content Forensics

With the rapid development of computer network technology and the extensive applications of high definition digital camera and image editing software, a variety of digital photos can be acquired much easier than ever before. Particullarly, it is much easier and faster to modify pohtos with the aid of the mature application of image editing software.Though new technology can bring many conveniences, it also brings much negative impact to our work and life. When we appreciate the wonderful images brought by digital photography, we are also plagued by some "fake" photos. Moreover, the forged digital pictures in the network is intensified currently, these fake photos have brought bad influence to people’s work and life, even brought a great negative impact to the society, which make people think "seeing" not necessarily "for real". Therefore, it is of theoretical significance and practical relevance to investigate the problem of the digital image forensics.Digital image content forensics includes two classes of research directions: active forensics and passive forensics. The active forensics is that the digital watermarking technology requires the watermarking information embedded with strong robustness, and thus can resist the attack and cannot be removed. For the passive forensics technology, most blind digital image forensics algorithms are true under many constraints, and thus the rate of accuracy, precision and practicability are much worser than the digital practical requirements. For the requirement of the robustness in the digital watermarking algorithm, this paper studies the problem of the digital watermarking algorithms under the framework of quaternion theory and QSVD. When considering the requirements of the accuracy, precision and practicability of the blind digital image forensics algorithm, under the extended block sort, this paper is concerned with the problem of forensics technique based on the copy-move operation. Main work and contributions of this paper are described as follows:1. Propose a novel blind watermarking scheme based on quaternion and jointed QSVD blocksFor the digital watermarking algorithm in color image, most of the existing results did not use the complete information of a full color image, and thus cannot reflect the relationship among the color channels. In order to completely utilize the information of color image and reflect the color channel connection, the quaternion model is applied to the digital watermarking algorithm. The proposed method in this paper tries to insert the watermark in the joint QSVD blocks where the color image is considered as an array of pure quaternion numbers. Then, the array of pure quaternion is divided into non-overlapping blocks and performs QSVD to the blocks, and the watermarking is inserted into the S blocks by changing the 2σ of S of the two adjacent blocks. Experiments shows that the proposed method can resist common attacks and better meet the requirements of robustness.2. Propose a blind watermarking scheme based on U matrix through QSVD transformationBased on QSVD transformation, most of the existing watermarking algorithms embed the watermarking information to the original color image through moderating the maximum value of the S matrix, but ignore the other two unitary matrices U and V. Since the energy of color image is not only related to the singular value, but also be related to the unitary matrix achieved by QSVD transformation, this paper resorts to change the first column of U matrix through QSVD decomposition, and then analyzes the impact on the image. Under this condition, one can choose the most sensitive coefficient to imbed the watermarking information, and thus realize the blind watermarking algorithm. Moreover, the proposed algorithm can satisfy the requirements of robustness and transparency.3. Propose a novel algorithm for image copy-move forgery detection and localization based on SVD and Projection DataFor the image forensics algorithms using copy-move operation, Most of the existing algorithms are designed by comparing the the feature of blocks pixel by pixel to detect the faked image. The disadvantages of these existing image forensics algorithms are the large complexity and long operation time. Based on the S matrix elements achieved by SVD transformation, This paper proposes a new algorithm in the framework of extended block sorting that can firstly filter out more independent blocks. Then, the second filtration is carried out to filter out more blocks using the projection data of the horizontal direction and the vertical direction, which realizes to detect the forged image caused by copy-move operation. The experiment results show that the proposed algorithm not only effectively detects multiple areas of copy-move forgery and precisely locates the duplicated areas, but also has stronger robustness for Gaussian blurring, JPEG compression and their mixed operations.4. Propose an efficient algorithm for image copy-move forgery detection based on DWT and Mean operationIn order to further reduce the complexity of the image forensics algorithm for copy-move operation, this paper proposes a new method based on the expanded block, where the one-level DWT is firstly applied to the image, and LL sub-image is obtained.In this case, the LL sub-image is divided into overlapping blocks with certain size, and thus more independent blocks are filtered out through the S matrix elements achieved by SVD transformation. Then, using the new blocks achieved by the mean operation on the reservation of the blocks after the first filtration, one can obtain some important characteristics by applying SVD transformation to the new matrix, which yields that the second filtration are carried out to filter out more blocks. This method can realize to detect the forged image caused by copy-move operation and further minimize the complexity of the algorithm with fewer features to represent each block. Through the experimental analysis, the proposed algorithm can effectively detect multiple areas of copy-move forgery and precisely locate the duplicated areas, and resist the common attack operation.