Copy-Move Image Forensics Alogrithm Based On Points And Sector Mean

With the rapid growth of human demand for information, digital image which is one of the information carriers has been widely used in various fields. At the same time, image editing software makes the image tampering much easier, and it can’t be determined by the human eyes. If tampered images are used in some formal occasions, such as the court and medical, it would undoubtedly cause a series of confidence issues. Therefore, the authenticity of digital images should be protected. At present, the identification technology which is used for digital images’authenticity has been paid closed attention in the domestic and overseas. In this thesis, the research content is image blind forensics technology based on keypoints and block-matching.The thesis firstly analyzes the current status of digital image forensics technology. This thesis deeply studies and analyzes the image forensics algorithms based on keypoints and block-matching. It also summarizes the advantages and disadvantages of the existing algorithms.Simulation of the copy-move image detection algorithm based on Harris corner points and step sector statics is performed. Tampered images of hiding targets and adding targets are detected and analyzed. Some detection problems of the algorithm are found. It can’t detect the forgery of flat regions and small regions. It also fails in detecting the forgery of multi-target pasted. This thesis firstly discusses the factors which affect the extraction of Harris corner points. It also analyzes the influence of the sector radius on the security of the algorithm. Then, a copy-move image detection algorithm based on Harris corner and sector mean is designed. To effectively improve the ability against the attack of removing objects with flat regions, this thesis needs to extract enough keypoints in the flat regions by changing the corner response function and reducing the contrast threshold. The designed algorithm improves the detection rate of small tamped regions by selecting the proper radius of the sector. It also improves the detection rate of multi-target pasted tamper by using G2NN algorithm. At last, it removes false detection through RANSAC algorithm. Experimental results indicate that the designed algorithm could detect the forgery of smooth regions and small regions. It can also detect the forgery of multi-target pasted. However, there are some false detection problems in the algorithm. It fails in detecting the forgery of large scale.To detect the forgery of large scale and reduce the false detection rate, an algorithm based on SIFT keypoints and sector mean is designed. Firstly, SIFT keypoints are extracted in the image. Then, regional descriptors which consists of 36 sector means are developed to represent every small circle image region around each SIFT keypoint. It matches the features according to the G2NN algorithm. It distinguishes different cloning regions through clustering. It reduces false detection by RANSAC algorithm and ZNCC algorithm. Finally, it locates the tampered regions accurately. Simulation results indicate that the algorithm could resist a certain zoom scale attack, and it could accurately locate and mark the tampered regions.Lastly, a simulation system for these algorithms is completed, and the simulation results are analyzed.