Study On Digital Image Copy-Move Forgery Passive Forensics

With the rapid development of science and technology,various digital products have brought great conveniences to human life.People can get digital images easily than ever before.Moreover,it is easy to use more and more digital image editing software to process digital images.However,one can also use these digital image editing software to tamper digital images for their particular desires.Digital image copy-move forgery is a typical image forgery.Tampers usually copy and paste image regions in a same digital image to hide or enhance its sensitive features.However,it will falsify the authenticity and against the objective facts of this digital image.Moreover,the personal property,reputation,and even national security may also be impacted.Therefore,it is very necessary to detect the authenticity of digital images in practical applications.The studies on passive detection schemes for digital image copy-move forgery gradually form an important research field.In this thesis,we focus on the studies of passive detection schemes for digital image copy-move forgery.The execution efficiency of the proposed schemes and the robustness for resisting more post trace hidden attacks are also considered.The primary contributions are list as follows.1.In order to detect and locate the tampered areas for digital image copy-move forgery and improve the execution efficiency of passive detection algorithms,a passive authentication scheme for digital image copy-move forgery is proposed based on a package clustering algorithm.The proposed scheme uses the means of the four quadrants in DCT coefficients of each image block to construct a feature vector to represent this image block.The detection accuracy of the proposed scheme is improved by using a package clustering algorithm to replace the traditional lexicographic order algorithms.The experimental results show that using the means of the four quadrants in DCT coefficients to construct a feature vector to represent this image block can reduce the feature vector dimension and improve the matching efficiency of the proposed scheme.In the similarity matching stage,all feature vectors of image blocks are clustered using the proposed package clustering algorithm.All feature vectors of each package and its adjacent package will be matched.It can reduce the matching times of all feature vectors.The proposed scheme not only can detect and locate multiple tamper areas,but also can resist some post trace hidden operations,such as adding white Gaussian noise,Gaussian blur,and other hybrid post-processing operations.2.To further enhance the robustness of resisting more complex post trace hidden opera-tions and improve the accuracy of the proposed schemes,a passive authentication scheme is proposed to identify duplicated regions for copy-move forgery based on perceptual hashing algorithms.The proposed scheme can improve the detection accuracy by using the perceptual hashing coefficient of each image block to represent itself.The experimental results show that the proposed scheme not only can improve the matching efficiency of the proposed algorithm,but also can detect the approximate image pixel distributions by using the mapping relation of perceptual hashing algorithms for image pixels.Moreover,a quadratic package clustering algorithm is proposed to further improve the precision of the proposed scheme.Feature vector matching algorithm firstly clusters the feature vectors of all image blocks and then clusters the feature vectors again in each package with their perceptual hashing coefficients.It can further enhance the precision of the proposed matching algorithm.The proposed scheme can resist more post trace hidden operations,such as adding white Gaussian noise,Gaussian blurring,adjusting contrast ratio,luminance,hue,and their hybrid operations.3.To consider the effective detection for a special image region tampering method that the copied regions will be rotated before they are pasted,a passive authentication scheme is proposed to identify duplicated regions for image copy-move forgery based on the rotational invariant features of the FAST feature points.In the proposed scheme,the FAST feature points are used as the main body to instead of image blocks.The rotational invariant feature vectors are constructed with all FAST feature points as the target points.The purpose is to reduce the number of rotational invariant feature vectors to improve the overall efficiency of the proposed scheme.All rotational invariant features are extracted by drawing circles and rings at the FAST feature points of images.The experimental results show that the constructed rotational invariant feature vectors can correctly represent the image areas that are rotated with any angle before they are pasted.It can enhance the robustness of the proposed scheme for multiple forgery methods in digital images.In the proposed feature vector matching algorithm,the efficiency of cross-matching large number of rotational invariant feature vectors is improved by using a reversed-generalized 2 nearest-neighbor algorithm.