Research On The New Algorithms Of Region Copy-move Passive Forensics Of Digital Image

The development of the network makes digital image very easily to be spreaded andshared, and the popularity of image processing tools makes digital images can be easilymanuplicated. If the forged digital images are used in scientific discovery, court evidence ornews reports and so on, it will have negative influence to social and political stability.Therefore, the research of digital image forensics has important significance for ensuringpublic order, cracking down on crimes and maintaining judicial justice and news integrity.This thesis is focused on the research of passive blind image regional copy-moveforensics technology. Firstly, we elaborate the background and the state-of-art of digital imageforensic technology, then analyze the deficiency of existing forensics algorithms, andsummarize the main jobs of this thesis. After that, related theory knowledges such as wavelettheory, Fourier transformation theory, and phase correlation technologies are introduced.Finally, we put forward two passive blind image forensics schemes as below:1. By using localized time(space) frequency analysis in wavelet domain and theprominent peak of phase correlation, an effective image regional copy-move forensics schemebased on phase correlation is proposed. Firstly,1level wavelet decomposition is done to agiven image, and then six dimensions of characteristic are extracted from each block in the lowfrequency sub-band coefficient covered by a sliding window. Finally, the duplicated region canbe detected through the dictionary sorting and phase correlation. It has a good balance betweendetection rate and false detection rate, and it is robust to resist post-processing attacks, such asgaussian blur, gaussian white noise, and severe JPEG lossy compression.2. Different from previous algorithms which need a huge image database, the proposedmethod can efficiently detect image regional copy-move from one image. Firstly, a colorimage is converted to grayscale image, then sensor pattern noise is extracted by usingdenoising filter in wavelet domain, after that, the variance of sensor pattern noise, SNRbetween denoised image and sensor pattern noise, information entropy and average energygradient from the original grayscale image are selected as the characteristics for forensics.Finally, by sliding non-overlapping window over the image, the tamper can be detected in thegiven image through analyzing the related degree between sub-block and integral image. Thealgorithm has high detection rate, and has a good robustness against bluring, adding noise,severe JPEG compression, arbitrary angle rotation, scaling and so on.The proposed two algorithms have high detection rate and high robustness for a single image, and they have a certain promoting significance for the research of digital imageforensics.