| In the rapid development of multimedia technology today,digital images have become an important carrier of visual information.However,with the various image editing software functions are increasingly perfect,many digital images have been tampered and spread in the network,causing the authenticity and integrity of the image to be destroyed.In this context,digital image forgery detection theory and method research has been generally concerned.Copy-Move Forgery Detection(CMFD)is one of the hot spots in the field of digital image forgery detection.However,CMFD technology has the disadvantages of low detection precision and high complexity.In order to obtain high precision effect of forgery detection,this paper adopts different theories and methods to carry out the following three aspects around the local region image forgery and has made certain research results:1.An adaptive CMFD algorithm based on feature matching of SURF(Speeded-Up Robust Features)is proposed,which solves the problem that the existing methods can not effectively detect the tampering of smooth region or small region.This method extracts the probability density SURF keypoints adaptively combined with the idea of superpixel segmentation and image information entropy classification,so that the keypoints of the entire image are evenly distributed.Then,the exponent moments with strong robustness are extracted in the circular feature region centered on each keypoint.The BBF(Best Bin First)and the Reversed generalized 2 Nearest Neighbor(Rg2NN)algorithm are used to perform fast multi-matching,using random sample consensus,zero mean normalized cross-correlation and other methods to locate and mark the tampering region accurately.The experimental results show that the algorithm has high detection accuracy and strong robustness.2.A robust CMFD algorithm based on multi-granularity superpixel matching is proposed.The algorithm uses “coarse-granularity” superpixel to determine the suspicious region,and then uses “fine-granularity” superpixel to precision positioning.Firstly,we use the method of entropy rate superpixel segmentation to segment the image.The features of the superpixel are represented by the keypoints and the quaternion exponent moments,and the keypoints are extracted from the improved Scale-Invariant Feature detector with Error Resilience(SIFER)by the color invariant model.The Exact Euclidean Local Sensitive Hash(E2LSH)algorithm matches the keypoints to determine the matching coarse-granularity superpixels.Then,the fine-granularity superpixels substitutes the matching keypoints to markthe suspect region.Finally,the adjacent similar superpixels are merged and the morphological processing is performed on the merged regions to obtain accurate positioning.Compared with other detection methods,the algorithm has good performance,which not only has high detection accuracy,but also has strong robustness.3.A CSH fast matching CMFD algorithm based on dimension reduction feature is proposed,which solves the problem of excessive time complexity on traditional block-based detection methods while maintains extremely high precision.The algorithm divides the image into global overlapping patches,extracts the features using nonsubsampled shearlet transform and reduces the time and space complexity by Singular Value Decomposition(SVD)to ensure the timeliness of the scheme.Then,a new CSH matching method is used to quickly match the global patch between images.Finally,the maximum likelihood method is employed to estimate the geometric transformation parameters and the tampered regions are marked with the optimized fast Zero mean Normalized Cross-Correlation(ZNCC)algorithm and morphological operation.The experimental results show that the algorithm has obvious advantages in accuracy and timeliness. |
PDF Full Text Request