Research On Color Image Copy-Move Forgery Localization Algorithms Based On Quaternion Transform

In recent years,with the rapid development of digital image processing software and Internet technology,the cost of forging and spreading digital images becomes lower and lower.The emergence of the forged images brings enormous threats to justice,criminal investigation,reputation protection and even social stability.Therefore,the study of effective image forgery localization algorithm becomes more and more important.The existing color image forgery localization algorithms usually process the three channels of color image respectively,thus ignore the holistic property of the three channels and the inherent correlation between them.Therefore,in this thesis,Fractional Cosine Transform(FrCT)and Fractional Zernike Moments(FrZMs)are effectively generalized to color image processing by using the quaternion holistic manner,and then these generalized transforms are combined with an improved PatchMatch algorithm for color image forgery localization.The main research contents of this thesis are as follows:(1)By using quaternion color image representation method,Fractional Quaternion Cosine Transform(FrQCT)and Fractional Quaternion Zernike Moments(FrQZMs)are defined to generalize the traditional FrCT and FrZMs respectively to quaternion field.At the same time,by deriving the relationships between FrQCT and FrCT,as well as between FrQZMs and FrZMs,the effective discretization implementation methods are proposed for FrQCT and FrQZMs respectively.(2)The FrQCT is used to extracted features from color forged images.An improved PatchMatch algorithm using reflection offsets and priority-based matching strategy is considered to match the similar features.Then,an effective color image forgery localization algorithm is proposed.The experimental results on two public datasets(FAU and GRIP dataset)have shown that the proposed algorithm outperforms other existing algorithms,and is with high robustness against additional operations and attacks.(3)The improved PatchMatch algorithm used in(2)is affected by its random initialization greatly.So,a Reliable Order-Statistics based Approximate Nearest Neighbor search Algorithm(ROSANNA)is considered to improve the random initialization and neighborhood propagation in the improved PatchMatch algorithm.Then,FrQZMs is used to extract the features from color forged images,and a new and more effective forgery localization algorithm is proposed.The experimental results on the two public datasets have demonstrated that the proposed algorithm improves the previous algorithm in(2),achieving a better performance.