Forensics Of Copy-Move Forgery For Digital Images

In recent years, with the development of digital media technology, the digital image has become a major information carrier. However, the popularity of image manipulation tools make it easy for people to tamper with the digital image. If these tampered images are maliciously applied to court forensic, news reports and academic research, the social credibility crisis will be arose. In this context, the digital image forensics technology arises at the historic moment. The technology can accurately determine the authenticity of images by analyzing the inherent characteristics in them. Copy-move forgery is a common way of image manipulation. It refers to copying a region in the image and pasting it on the other areas in the same image. This paper mainly focuses on the blind forensics of copy-move forgery. Due to the copy-move operation will introduce at least two high-similarity regions in image, the detection method can take the similarity of image features as criterion to find the copy-move matching areas. The specific research results are as follows:(1) A copy-move detection algorithm based on Zernike moments is proposed. The Zernike moments which are extracted from overlapping image blocks are used for constructing the feature vectors. In the feature matching procedure, a new matching method based on color partition and locality sensitive hashing is proposed. The process of color-based grouping and hash mapping reduces the amount of calculation. And the matching accuracy is improved by using the relative distance instead of Euclidean distance as the similarity criterion. In the error reduction procedure, the false matches are removed by the proposed clustering-based isolated block elimination method and the RANSAC algorithm based on slope clustering. This error reduction mechanism can not only effectively remove the false matches, but also retain the matches from different copy-move groups well, which guarantees the multi-copy-move regions can be fully detected. The experiment results show that the algorithm can effectively determine whether the tested image has been tampered and localize the duplicated regions precisely. In terms of robustness, the algorithm can detect the smooth duplicated regions and has a good performance on resisting rotation, slightly scaling, JPEG compression, blurring and mild noise adding.(2) To deal with the problem that the keypoint-based copy-move detection method fails to detect the smooth duplicated regions in image, a BRISK-based adaptive copy-move detection algorithm is proposed in this paper. Firstly, a tested image is divided into smooth regions and texture regions by the texture distribution in the image. Because the BRISK keypoints are the extreme points in the scale space, they often appear in the texture regions. In order to acquire more keypoints in the smooth regions, the algorithm reduce the keypoint extraction threshold in these smooth regions. And in the texture region, the algorithm still use the common extraction threshold, which can avoid the extracted keypoints in texture regions to be too dense and increasing the amount of calculation in the algorithm. Due to the BRISK descriptor is512D binary sequence, the algorithm use hamming distance as a similarity criterion. It is worth noting that the computational complexity, of hamming distance is much lower than that of Euclidean distance, so this algorithm has better real-time performance compared to the traditional algorithms. Experiments show that this algorithm can effectively detect smooth copy-move regions, and is robustness against rotation, scaling, mild JPEG compression, blurring and slightly noise adding.