Research On Blind Forensics Technology Of Digital Image Copy-move

With the popularity of digital image acquisition devices, digital image is widely used in various fields of the society. However, with a variety of image processing software developing, the digital image can be easily tampered. More and more forged digital image have appeared in our life and bring serious adverse effects to the human society. Thus the society demanding for blind forensics technology is gradually highlighted.Blind digital image forensics is to assess the authenticity of a digital image without any prior information. The research on blind forensics technology of image copy-move is high-profile in image forensics research field. Copy-move tampering is usually referred to the image manipulation That a small region of any image is copied and then is pasted to another region of the same image. As the copied area is from the same image, it has better concealment and is not easy to be perceived. Copy-move tampering is easy to operate, so it is widely used. For the forensics of image copy-move tampering, although there has been many professionals putting forward to their own way, there are still many challenges.This paper analyzes the existing typical approaches of copy-move tampering detection, starting as solving a series of problems such as the high matching computational complexity in the general algorithm, the multiple copy-move tampered region detection and low accuracy for locating, we analysis the characteristics of SIFT algorithm qualitatively, test the robustness of SIFT feature and order research on the detection algorithm of digital image copy-move tampering with SIFT feature. The test dataset is produced used Photoshop. The copy region and pasted region are selected randomly to remove the impact of the algorithm by the image map.The main research achievements achieved in this thesis include the following three aspects:Firstly, an improved method based on SIFT is proposed for blind detection of image copy-move tampering. An effective feature matching method-ng2NN(new generalization 2 Nearest Neighbor) is proposed and the fast ZNCC(Zero Mean Normalized Cross Correlation) trategy is introduced in the tampered localization. Ng2 NN is improvement of g2 NN. In the process of ng2 NN matching, the keypoints are divided into half recursively firstly and the keypoints in every combination of two half are matching with ng2 NN.In the detection algorithm, the SIFT feature is extracted and matched with ng2 NN firstly. Then the matched keypoints are clustered with the J-Linkage clustering algorithm. The affine transformation model between the copy block and the transformed image is estimated according to the clustering results. The test image is transformed with affine model. The correlation map between the test image and the transformed image is estimated with fast ZNCC strategy and the tampered region is located quickly and accurately based on the correlation information finally. From the experimental results, we can see the efficiency of the feature matching and the tampered region location of the improved algorithm. The proposed algorithm can provide the better position effect even the multiple copy-move.Secondly, an anti-flip algorithm of image copy-move forgery detection is proposed. Most existing methods of copy-move tampering detection do not consider the flip transformation. The image preprocessing is added into the proposed method. Using SIFT feature the proposed method can not only detect the general copy-move forgery, but also the deceptive stronger copy-flip-move forgery. The algorithm can resist on rotation and zoom. In addition it is effective for the multiple copy-move forgeries.