Copy-Move Detection Algorithm Based Onimage Segmentation

With the development of image editing technology,images are easily tampered.These tampered images can easily cause the authenticity of the image to be questioned,which may have a great influence and break on the society.Therefore,forensic technology is needed to protect the authenticity of the image and identify the authenticity of the image.This thesis firstly introduces the background and significance of digital image forensics technology,and summarizes the existing passive forensics techniques of digital images.Aiming at the research of region duplication image forgery,this thesis analyzes the existing forensics algorithms based on image block and feature point,and it is pointed out that it is difficult to detect the tamper in the smooth region and can't resist rotation attack.Aiming at the problem of detecting tampering in a smooth area and resisting to rotation,a copy-move detection algorithm based on super pixel segmentation and pixel clustering is proposed,SLIC super pixel segmentation algorithm and clustering are combined to divide the image into two classes,one is the rich texture region,one is smooth region,clustering is based on the mean and standard deviation of pixels in the super pixel segmentation block.By extracting the dense Harris points,the rotation invariance sector mean feature is obtained.The G2 NN algorithm and the RANSAC algorithm are used to match the features and remove mismatches.Compared with the literature algorithm,the experimental results show that the proposed algorithm has higher true positive rate and lower false positive rate for fuzzy,noise adding,JPEG compression and rotation attacks,it also can detect tampering in the smooth area.Finally,the influence of the initial parameters of SLIC super pixel segmentation block on the performance of the algorithm is analyzed.In order to enhance the performance of the algorithm,the clustering method based on SLIC super pixel segmentation has been improved,the improved clustering method is based on the proportion of the number of SIFT feature points in the segmented block in the segmented block as the basis of clustering analysis,and it uses this basis for k-means clustering,after classification,the rich texture area and the smooth area are obtained.SIFT features and the sector mean features are used as matching features in rich texture regions and smooth regions respectively,and then G2 NN algorithm and RANSAC algorithm are used to perform feature matching and remove mismatch.Compared with related literatures,the improved algorithm has better detection performance.