Blind Identification And Location Of Image Copy-move Forgery Based On Regional Features

With the development and popularization of computer image processing technology,image forgery has become a ubiquitous behavior in people's life,even in scenarios such as news reports and academic researches that have high requirements on image authenticity.The behavior has serious negative effects on fairness and justice of society.Therefore,an effective identification of tampering has become a priority to be achieved.This paper studies the most common method in image forgery—copy-move,and researches the identification and area positioning methods of ghost-like copy-move forgery occuring in texture-intensive areas and concealed copy-move forgery occuring in smooth areas.(1)The methods of idetifying and locating the forgery in the dense texture ares are studied.Considering the copy-move forgery occuring in texture-dense areas,common feature-point based identification methods exclusively consider the gray information of image,which lead the high mismatch rate and the inaccurate clustering results.This paper extracts the color feature point information from the image,proposes a bucket-based g2nn matching strategy,which significantly reduces the time consumption of feature point matching and clusters the matched feature points in the affine transformation matrix dimension to effectively solve the problem of classification of feature points when the distance between the tampering source area and the destination is close.Finally,the areas which have feature pointsis are locally searched expansively and in this way,tampering areas can be located more completely.Experiments show that the method effectively reduces the mismatch of feature point pairs and is robust to a variety of tampering additional operations.It not only can identify whether the image has been tampered with,but also can determine the tampering area more specificaly.Compared with the common feature point algorithm,the image tampering identification accuracy and recall rate of the algorithm are increased by 4 and 6 percentage points on average on the same data set.(2)The methods of idetifying and locating the concealed forgery in the smooth ares are studied.The concealed copy-move forgery generally occurs in the smooth areas of the image,and the number of feature points in these areas is not enough to detect tampering when using the method based on feature-points.This paper proposes a block-based detection method that combines image moment features and texture features.It extracts Zernike moments from image blocks and SVD values after rotation-invariant LBP encoding,it also establishes k-d tree for searching the nearest neighbor of each feature and determines the similarity of both features combing the features and the distance between corresponding blocks.Experimental results show that the method is robust to JPEG compression,Gaussian white noise,and image blur.Compared with the single feature,the joint feature is more recognizable and can better perform tampering identification and location in smooth areas.Compared with general block feature detection methods,the image tampering identification accuracy and recall rate of the algorithm are increased by 6 and 8 percentage points on average on the same data set and there is a nearly 10-20 percentage point improvement in the average accuracy of regional positioning.This paper proposes two methods of identification and localization for different types of copy-move forgery.Experiments show that the boh methods have a certain degree of robustness,high accuracy of detection and high completeness of regional locating in the case of multiple tampering additional operations.