Research On Image Region Duplication Forgery Detection

With the rapid development of varieties of image editing softwares and digitalimaging devices, the number of digital image forgery is steadily on the increase, which hasposed a severe negative impact on image information security in different imageapplication areas. Therefore, it has become an important problem to effectively regulate thebehavior of image forgery and timely detect image forgery in the field of imageinformation security. Currently, region duplication forgery is one of the most commonlocal tampering ways, it is for the same image that copy one part of image and paste intothe other disjoint region so as to achieve the purpose that hide or cover part of the image.Thus this operation will change the meaning of original image.Passive region duplication forgery detection is not only one of the most importantbranches in image forensics, but also is a research hotspot in image information securityfield recently. In this paper, the research direction is mainly for image region duplicationforgery detection, which mainly consists of four parts. Firstly, research of circular windownormalization, scanning the image with a circle window and taking the polar coordinatestransform to expand the circular window image, thus a rectangle image is generated by thebilinear interpolation finally. Secondly, research of adaptive frequency band limitationbased on fractal dimension, taking use of the fractal dimension method to estimateroughness features in the circular window image, and design a piecewise function betweenfractal dimensions and limitation coefficient, so as to achieve adaptive frequency bandlimitation, which can efficiently reduce the impact on the detection algorithm'sperformance by noise, blur and other high-frequency informations. Thirdly, research ofoffset angle estimation based on standard phase correlation. Adopt the standard phasecorrelation method to estimate the rotate angle between two matching circle windowimages. Finally, research of forgery region boundary estimation, taking use of the seedfilling algorithm and the estimated angel to identify the copy region and pasted region, andset the seed by two original matching circle window images, continually search the similar region blocks recursively within the four neighborhood of the seed. Until the peak value isless than similarity threshold or the block has been searched, the searching procedure isstopped, and all successfully matching blocks are marked with a special color for displayto users.On the basis of experimental data, the proposed method can effectively improve theregion duplication forgery detection accuracy, also has a good detect performance for themixed post processing retouching operation, such as Gauss noise, usual geometricalrotation, flip (horizontal flip and vertical flip), JPEG lossy compression, illuminationchange.