Research On Blind Identification And Localization Algorithm Of Copy-move Forgery Image Based On Keypoint

With the advent of digital era and the development of information technology,digital image has gradually replaced traditional film image due to its fast and inexpensive advantages and a good balance between imaging effect and cost.In order to meet different needs and facilitate the processing of images,various image processing tools appear.The software is constantly improved,making the tampering of the image easier.These covert ways of forgery images,once used by some criminals with ulterior motives,will bring many adverse effects.For example,they seriously destroying the authenticity of news reports and affecting the fairness of judicial evidence.In addition to strengthening industry self-discipline,making relevant laws and regulations and improving people's ability to distinguish images,the present solution can also identify these forgery images by digital image forensics.Digital image forensics technology has two kinds of active forensics and blind forensics,which mainly determine the authenticity of the image,the way of tampering and the area of tampering by means of technology.The technology of active forensics identifies the authenticity of images by judging the integrity of the prior information such as digital watermarks or digital signatures that are added to the images in advance.However,the blind forensics technique omits the step of adding prior information and can identify forgery image by analyzing and extracting the image features.Therefore,the technology of active forensics is more limited,and the application of blind forensics is more extensive and more realistic.This paper mainly studies one of the mainstream ways of the blind forensics-copy-move forgery image blind identification technology.First,the background,research status and necessity of digital image forensics are introduced.The development history of copy-move forgery image blind forensics is emphatically introduced among them.Then,according to the identification algorithm is based on image blocks,keypoints or the combination of image blocks and keypoints,several classic blind identification algorithms are introduced respectively,and their advantages and disadvantages are summarized.Finally,based on the defects of feature descriptors used by the existing algorithm and the shortcomings of the clustering algorithm,two improved copy-move blind identification algorithms are proposed based on keypoints.Aiming at the problems of high computation complexity,low accuracy and inaccurate tampered region location,a improved copy-move forgery detection method using a new feature descriptor-LATCH is proposed in this paper.First,keypoints are extracted by the classical SIFT and LATCH features are described for corresponding keypoints.Then LATCH features are matched by using the hamming distance.Subsequently,remove false matching by K-means clustering and estimation of geometric transform parameters.Finally,in order to locate tampered region accurately,a new recursive method based on region-like growing is proposed.Experimental results confirm that the proposed method is not only effective for geometric transformation and robust to post-processing,but also has higher accuracy on tampered region location and less time consumption.Besides,it has great performance on the type of hiding object forgery.Aiming at the situations that the past copy-move forgery detection algorithm can't handle multiple copy-move forgery and can't locate forgery region accurately,in addition,the hierarchical clustering algorithm can't find non-convex clustering,a new copy-move forgery detection method is proposed in this paper which based on the feature of AKAZE and DBSCAN of the density clustering methods.First,keypoints and corresponding features of the image are extracted by AKAZE method.Second,features are matched by using the Hamming distance and G2 NN which is the way of multiple matching.Then,the DBSCAN is used to cluster the keypoints and remove false matching.Finally,in order to locate forgery area accurately,the method of peak signal to noise ratio is used to find the most suitable filling radius.Experimental results show that the proposed method performs well both for geometric transformation and post-processing as well as multiple copy-move forgery,it and can locate the forgery area accurately.