Research On Copy-move Forgery Of Image Blind Forensics

With the promotion and application of various types of image processing software,the number of tampered images is increasing, which seriously affects the imageinformation security in different industries. Therefore, at present image forensics hasbecome an important issue. Image copy-move forgery tampers on certain key regions orimportant objects in one image. Tampering traces are difficult to be observed by eyesbecause of similar characteristics of color, brightness information. Therefore, copy-movemanipulation is widely used in image forgery.Most existing copy-move forgery of image blind forensics algorithms are vulnerableto mixed distortions. This paper presents a novel and effective image blind forensicsmethod that resists to retouching, lossless compression of JPEG and geometric distortions.The main contents include image scanning and circle window expanding, image matching,offsets calculating and accuracy analyzing and tampered region searching. First of all, acolor image should be transformed into a gray image in the preprocessing, and the grayimage can be scanned by1step length. This scanning mode possesses favorable rotationalinvariance.The divided circle window should be transformed into Log-polar coordinatespace which is invariant to rotation and scaling. The way can convert rotation and scalinginto translation effectively. Then circle window radius can be adaptively adjusted to keepthe content consistency in two circle windows, which can resist to strong scaling attacking.The relationship between fractal dimension and characteristic frequency can be constructedto achieve shrinkage coefficient, which can adjust the width of frequency bandautomatically, and this method can eliminate noise effectively. The combination of the twotechnologies can improve the peak of phase correlation in order to match images whichinvolve noise, rotation and scaling attacking. There for rotation angle and scaling factorcan be calculated referring to the peak position of the phase correlation matrix whenautomatic adjustment circle window images are the input images. Finally, these offsets areapplied to region searching, and copied regions and tampered regions can be located. Experimental results show that the proposed method, which has high accuracy rateand low false detection rate, can resist single or mixed type attack which include gausswhite noise, JPEG compression, translation, flipping, rotation and scaling toleranceeffectively.