Image Forensics Research On Copy-Move Forgery

With the development of computer technology, the application range of digital images has become very wide. However, a variety of image processing software has made digital images easily been tampered, which bring many negative impacts for the community. Therefore, appropriate technology is urgently needed to detect the authenticity of the digital image.At present, the digital image forensics can be divided into active forensics and blind forensics. For active forensics, preprocessing from image providers is needed, while blind forensics is independent on any preprocessing. Therefore, blind forensics has become the present focus of digital image forensics research. This thesis researched for the blind forensics of regional copy-move forged image. To cover the original information, the regional copy-move forgery operator copies certain area of the image and pastes it into another area of the same image. This thesis finally proposed a novel algorithm for detection and location, the research results are as follows:Firstly, we proposed a new feature description algorithm called LPP-SIFT (Locality Preserving Projection-Scale Invariant Feature Transform). This descriptor can generate effective and low dimensional characteristics description for SIFT.Secondly, this thesis used LPP-SIFT characteristics descriptor in copy-move forgery detection and location for the first time. The descriptor can effectively describe the attributes of feature points, and its dimension is not high, which greatly improves the calculation speed of feature points matching.Thirdly, we divided the set of feature vector in order to match feature points, and we used the ratio of vector angle to judge if feature points are matched, if they do, segments are marked between them.Based on the proposed detection method in this thesis, we can judge a detecting image is copy-forged if there are gathered feature points, the gathered regions are copy-forged. Experiments demonstrate that the algorithm proposed in this thesis is effective for copy-move forgery detection and location, besides, it is greatly robust to ration, scaling, JPEG compression and noise addition.