The Research On Copying Forgery Of Digital Image Passive Forensics

Digital image has been used in news reports,security monitoring, scientificresearching and medical fields as a significant message carrier. However, with thedevelopment of image processing software which has friendly user interface andhigh performance ,digital image is vulnerable to a malicious tampering. So thedetection of digital image authentication has become extremely urgent.There are twomain directions in the field of image authentication detection. One is active forensicswhich is based on digital watermarking and blind forensics. This technology worksby inserting some prior information into the image for the authentication, so thismethod limits the development of researching. The other is blind forensics,thistechnology works on the information of the image itself, thus this method overcomesthe weakness of active forensics to a certain extent. The regional tampering is ausual way to tamper digital image, therefore it is very important to work out thedifficulties in this research field.This paper proposes two blind forensics schemesbased on two different tampered ways which include image splicing and copy-pasteforgery in the same image.(1) Image splicing is a normal way to tamper digital image.This paper proposesa blind and passive forensic scheme based on multi-features amalgamation to detectimage splicing.Firstly, we get three kinds of features of image by analyzing its phasecongruency and texture features, and disassembling it into the intrinsic modelfunction domain through bidimensional empirical mode decomposition.Utilizing thefeatures,we construct a forecast model with a support vector machine as theclassifier to judge whether the image is forged.Lastly,we evaluated the proposedscheme with the standard spliced image dataset.The experiment results indicate thatour scheme has higher detection with the feature vectors in a low-dimension.(2) Targeting the weak robustness of most existing algorithms in detecting thetampered image region with some post-processing operations, this paper proposes anovel region forgery detection algorithm based on point matching.Firstly,we extracte feature points with scale-invariant feature transform, and describe the points invirtue of principal component analysis. Owing to the similarity between the pastedregion and the copied region,we find all possible fogeries by seeking for similarpoint pairs by using their descriptors.With the robustness of the adopted method,ourproposed algorithm performed robustly in terms of not only additive noise, lossyJPEG compression and blur operation but also effective in rotating andzooming.Proved by tests, this method has good robustness to commonpost-processing techniques.