Research On Keypoint-Based Copy-Move Forgery Detection Algorithms For Digital Image

With the development of the Internet and the popularization of the portable devices.it becomes more convenient to obtain rich multimedia information.Along with continue mature of image editing software,the image content has been tampered frequently,including copy-move and splicing forgery.Events triggered by forged images are common in many fields such as news report and academic research,which seriously threaten the stability of society.Therefore,it is urgent to research the image forensics technology.Copy-move forgery(CMF)is the most common manner to tamper images,which refers to pasting part region of the image into another region of the same image to cover up the truth.The current researches are mainly divided into block-based and keypoint-based image copy-move forgery detection(CMFD)algorithms.The former has many problems,such as high computational complexity and difficult to resist large-scale geometric attacks.The latter has some problems,such as difficult to correctly judge whether the smooth area of the image has been tampered with,inaccurate location,and so on.This thesis researches the problems existing in keypoint-based CMFD algorithms.The main work done is as follows:(1)To solve the problem that most keypoint-based CMFD methods cannot correctly judge whether the tampering occurs in the smooth area of the image,an image CMFD method based on speeded-up robust feature(SURF)and accelerated-KAZE(A-KAZE)feature is designed.By combing SURF and A-KAZE detectors with smaller contrast thresholds,the keypoints are extracted to uniformly distribute throughout the whole image.In the stage of locating the forged regions,the rough areas of the duplicated regions in the tampered image are determined according to the dense points obtained in the feature matching.which reduces the scope of calculating the correlation coefficient map.An improved method of calculating the correlation coefficient map is designed,which is more conducive to the subsequent processing of locating the duplicated region.Experimental results demonstrate the designed method can locate the duplicated regions of plain CMF images,and can resist the small-scale rotation,scaling,JPEG(Joint Photographic Experts Group)compression,blurring,noise addition,and other operations.However,this method cannot locate the duplicated regions of the forged images in the smooth area with high brightness,and causes misjudgment in the images with too many similar regions.(2)In view of the fact that the previous method cannot correctly judge whether CMF occurs,a CMFD method based on SURF and polar complex exponential transform(PCET)is designed.To reduce the scope of feature matching,the image block classification based on the local entropy information is researched,which makes the duplicated reaions in the same smooth regions or texture regions.PCET coefficients are used as the keypoint feature instead of SURF descriptor.The mismatched keypoint pairs are eliminated by combing random sample consensus(RANSAC)algorithm and the method of counting the points in the image blocks.The rough areas of the duplicated regions are determined according the dense points obtained in the feature matching stage,in which block-based CMFD method and morphological operation are used to locate the duplicated regions.To solve the problem that the detected regions are smaller than truth duplicated regions,a strategy of expanding edge is adopted to expand the scope of detected duplicated regions.Experimental results demonstrate that the designed method can locate the duplicated regions when the CMF occurs in the smooth regions with high brightness and avoid the misjudgment in the images with too many similar regions.Besides,it has ability against rotation,scaling.JPEG compression,noise addition,and blurring.