The Research On Detection Of Image Region Copy-move Forgery

In today’s digital era, digital images that have become the information media are widelyused in various industries. The increasingly digital radiography make it easy to manipulatedigital images without leaving visible traces. Those tampered images may be widely used indifferent circumstances such as science, military, court and so on. As a result, spurious digitalevidences may cause great negative influence on the society and politics. Thus, to identify theauthenticity and integrity of the digital image has profound practical significance. Digitalimage forensics can be generally classified into active and passive forensics. The shortcomingof active forensics is that the digital signature or watermark must be pre-embedded in images.In contrast, passive forensics that authenticates digital images without relying on thepreprocessed information has a stronger applicability. Therefore, passive forensics hasbecome the main hot research topic in digital forensics field. On the basis of the researchesabout several invariant moments, this paper proposed two detection algorithms aiming atdigital image region copy-move forgery.A region copy-move forgery detection method resisting geometric attacks based onwavelet moment and statistical character was proposed. Firstly, the detected image ispreprocessed, it is divided into multiple overlapping blocks with the same size and mapped tothe polar coordinates. Wavelet moment invariants and some other additional statisticalcharacteristics of each block are fused to be the final feature vector. Secondly, the featurevectors from all blocks are sorted in alphabetical order and the tampered regions are locatedby combining the displacement vectors of image blocks. Experimental results have shownthat the proposed method performs well to deal with various post-processing manipulationsof image region copy-move forgery, including noise, blurring, rotation and so on.Another region copy-move forgery detection method based on Radial Harmonic FourierMoments was proposed. The basic idea is similar with the front one. The detected image isdivided into multiple overlapping blocks and the Radial Harmonic Fourier Moments of eachblock are extracted to be the final feature vector. Lastly, the tampered regions are detectedand located. In order to resist the noise effectively and reduce the time complexity of the algorithm, do the wavelet transform to the image before decomposing it, and then the lowfrequency part is chosen to be detected. Experimental results demonstrated that the proposedmethod can effectively resist the attacks of various post-processing manipulations, includingnoise, blurring, rotation, flip and so on. What’s more, this method narrowed the search spaceof similar block matching, reduced the time complexity and improved the detectionefficiency.