Research On Passive Forensics Methods Of Digital Image Copy-Move Forgery

With the increasing popularity of digital images in people's daily life and work,a large number of malicious tampered images have appeared in news reports,academic research,commercial activities,insurance claims,and other fields,which usually result in negative impacts on society.Therefore,how to carry out forensic analysis on the originality,authenticity and integrity of digital images has become an urgent problem to be solved.This paper focuses on the forensic methods for the most common copy-move forgery in digital image tampering.On the basis of the existing research results,we established a stepwise refinement mode for copy-move forgery detection,and carried out in-depth research on the three incremental stages of this mode:image-level forensic analysis,pixel-level forensic analysis,and quantitative assessment of detection regions.The main innovations and contributions of this paper are as follows:(1)The stepwise refinement mode for copy-move forgery detection is established.Taking into full account the process of copy-move forgery detection,and combined with the practical application requirements of digital image passive forensics,we propose to analyze the forgery image by a stepwise refinement manner.That is to say,according to the specific conditions of the detection image,we perform image-level forensic analysis,pixel-level forensic analysis,and quantitative assessment of detection regions in sequence,which is a process of qualitative forensic analysis to quantitative forensics analysis.The extensive experiments have shown that this forensics model can not only quickly exclude a large number of real images and provide quantitative credibility for security forensics analysts,but also help to uniformly optimize all detection parameters during copy-move forgery detection to achieve better detection performance.(2)The image-level forgery detection method based on UR-SIFT and two-stage feature matching is proposed.First,according to the distribution requirements of keypoints in copy-move forgery detection,we propose the keypoint detection algorithm based on UR-SIFT to achieve uniformly distributed keypoints in both image and scale spaces.Second,we propose the two-stage feature matching algorithm to obtain more correct matches.The experimental results show that the proposed method greatly improves the recall rate of tampered images,especially for test images containing smooth tampered regions,and has good robustness to Gaussian noise,JPEG compression,and geometric transformations,such as rotation and scaling,which builds a strong foundation for further quantitative forensic analysis.(3)The pixel-level forgery detection method based on improved match grouping and hybrid region localization is proposed.First,according to the characteristics of the copy-move forgery image,we propose the improved match grouping algorithm to effectively overcome the problem of the existing J-Linkage clustering,such as poor ability to resist mismatches and long clustering time.Second,we propose the block matching localization algorithm to significantly improve the detection performance of self-similar images.Finally,in order to make full use of the advantages of the correlation coefficient localization algorithm,the hybrid region localization algorithm is proposed.The experimental results show that the proposed method can outperforms other state-of-the-art methods in terms of both the detection performance and computing time.(4)Aiming at the problem that the keypoint matching stage and the correlation coefficient localization stage support the detection region is not completely consistent,we propose the tampered region assessment method based on evidence theory.We first quantify the impact of the keypoint matching stage and the correlation coefficient localization stage on the detection region,and determine their basic probability distribution function respectively.Second,we establish the assessment model for the detection region by the Dempster's combination rule,and calculate the credibility of the detection region.The experimental results show that the credibility of the detection region can well eliminate the erroneous detection region,and help the security forensics analyst make correct decisions.