Research On Forensic Method Of Copy-Move Forgery In Digital Image

Nowadays,the society is developing to the direction of informationization.Digital images are improving the way people communicate in daily life gradually.However,countless tampered images are made by powerful image editor.These forged images,which make the idea of believing photographs to be true seems unreliable nowadays,perturb the good social order.Therefore,the research of digital image forensic can make positive effect on fighting against crimes,maintaining journalistic integrity and judicial fairness.The thesis is focused on the detection of copy-move forgery in same image and the estimation of geometric transformation.Firstly,we elaborate the background and the state-of-art of relevant digital image forensic methods,and then analyze the existing forensics methods.Thereafter,we introduce related theories and knowledge of our method,such as scale-invariant feature transform(SIFT),principal component analysis(PCA),affine transformation,hierarchical cluster analysis and least median of squares(LMedS).And we expound our methods at last.The main achievements of our work consist of the following aspects:(1)We propose a method based on dimension-reduced SIFT feature to detect copy-move forgery.Our method makes use of PCA to reduce the dimension of feature vector in feature matching and justifies images' authenticity by the rule of keypoints'distribution in real image.Comparing with other methods,our method achieves a better tradeoff between true positive rate and false positive rate,and effectively reduces computational complexity.(2)We propose a method based on LMedS to achieve the estimation of geometric transformation.This method builds the best estimated model for geometric transformation based on the keypoints which extracted by image forensic method in this thesis.Test show that our method gets the accurate parameters of geometric transformation and needs not to set empirical threshold comparing with existing methods.Experiment shows that we can get the parameters of geometric transformation as well as maintain the balance between true positive rate and false positive rate through methods in this thesis.