Research On Passive Forensics Technology For Local Forgery Of Digital Image

With the rapid development of image editing software,people can forge digital images at very low cost.This poses a great threat to the dependability of the image.Copy-move forgery is a common operation in various image forgeries,in which one or more areas of an image are pasted to other locations in the same image to cover or replicate the objects of interest.The process may be accompanied by rotation,scaling,adding noise and compression to make counterfeiting more convincing.Generally,the detection of copy-move forgery is challenging.And existing methods have low detection performance in some case,such as copy-move forgery only involving small or smooth areas,or tampering areas subject to severe attack processing.To solve the above problems,this paper proposes three effective digital image copy-move forgery detection(CMFD)algorithms.The specific contents are as follows:1.Combining superpixel and Delaunay triangulation theory,this paper proposes an image CMFD algorithm based on integrated features of triangular grids.The method effectively solves the problems about high time complexity of block-based technology and low accuracy of keypoint-based technology.Firstly,combining the idea of super-pixel segmentation and image content classification,the color invariant Scale-Invariant Feature with Error Resilience(SIFER)detector is used to adaptively extract keypoints for different image regions.Secondly,a set of connected triangles is constructed on the generated points,and each triangle is described according to Fast Quaternion Radial Harmonic Fourier Moments(FQRHFMs)and gradient entropy.Thirdly,the Coherency Sensitive Hashing(CSH)algorithm is employed to quickly match triangles.Finally,Dense Linear Fitting(DLF)technique,optimized Zero mean Normalized Cross-Correlation(ZNCC)measure and morphological operation are adopted to locate duplicated regions.Experimental results show that the proposed algorithm improves the accuracy of traditional detection with reasonable memory and computation.2.According to the uniform measurement of keypoints and Fast Robust Invariant Feature(FRIF),this paper proposes an image CMFD algorithm based on hybrid descriptors of uniform keypoints.The method effectively solves the problem that the existing similar algorithms cannot detect non-significant visual objects.Firstly,a new adaptive feature point detector based on uniformity measurement is proposed.The detector inherits the advantages of block-based and keypoint-based methods,and combines uniformity measurement with adaptive iteration to extract stable points from each sub-block.Secondly,FRIF is used to construct mixed binary descriptors to represent the extracted points.Thirdly,the Reversed-generalized 2 Nearest-Neighbor(Rg2NN)technology is applied to find candidate matching pairs.And the correct pairs are queried by RANSAC after SLIC algorithm roughly divides the candidate pairs.Finally,the repeated parts are located and marked by using the optimized NNPROD method and morphological operation.A large number of experimental results prove that the proposed method is effective even under various challenging conditions.3.Combining the Geometry-Invariant Dense Adaptive Self-Correlation(GIDASC)local image feature descriptor,this paper proposes an image CMFD,identification and inpainting algorithm.The method can effectively locate the repeated regions while identifying the authenticity and repairing the falsified portion.Firstly,the iterative process of adaptive keypoints detector based on uniformity measurement is removed.The adopted method is to reduce the contrast threshold and adjust the image size,which improves the speed of extracting points.Secondly,GIDASC descriptor is used to represent the local image content.Thirdly,the features are matched and the candidate pairs are clustered using SLIC algorithm.And an iterative method is adopted to solve the problem that homography is not unique when there are multiple clones.Finally,the optimized NNPROD algorithm is applied to identify the authenticity of the repeated regions.The missing area is restored through the information of the adjacent area.The comprehensive results show that our method is superior to other advanced methods in CMFD and has unique innovation in forgery location and inpainting.