Research On The Blind Detection Of Forgery Images

With the rapid improvement of multimedia technology, devices and software of imageprocessing are used widely. So with them it is easy to process images for the desired resultsand purpose. However, the digital image is modified easily, which makes the imageinformation face a lot of safety troubles. If tampering image is exaggerated or distorted thetruth in scientific, political and military fields, this action is likely to have serious bad effectson economical development and social progress. Image authentication is a detectiontechnology which detects the originality, integrity and authenticity of images. The blinddetection technology of images has become a hot topic at home and abroad.In the thesis, blind detection algorithms of forgery images are researched. The main workis as follows:(1)A new detection algorithm of copy-move forgery images is proposed based on fractalsand statistics, in order to overcome the deficiencies that the most existing detection algorithmsof image copy-move forgery can not detect effectively sequential mixed image forgeries withregion duplication. The steps of the algorithm are as follows: First, it divides an image intooverlapping blocks, and the each block is respectively extracted to an eigenvector, which iscomposed by fractal dimension and three statistic data. Second, the eigenvectors of all theblocks are lexicographically sorted. Finally, the tampered regions are localized by means ofthe location information of the blocks and the Euclidean distance. This algorithm can detectnot only the traditional forgery images with copy-move operation, but also the multipleregional copy-move forgery images with additional operations of rotation, flipping, or theirmixture. Moreover, the proposed detection algorithm of forgery images has good robustnesswhen the forgery images are attacked by Gaussian blurring, contrast adjustment, brightnessadjustment, etc.(2)A fast detection algorithm of copy-move forgery images is presented by means ofreducing dimensions with the wavelet transform, in order to resolve the realtime problem oftraditional detection algorithms for forgery images with SIFT(scale invariant featuretransform). It is robust to forgery images undergoing some attacks like scaling, rotation, brightness adjustment, contrast adjustment, Gaussian blurring, etc. This new algorithm hasfaster matching speed and higher matching accuracy.The theoretical basis of the proposed two algorithms is respectively introduced in thethesis. Further, the detection effectiveness of the new algorithms is validated by theexperimental results and contrast results.