Analysis And Application In Digital Forensic Of The Fractal Features Of Digital Images

With the rapid development of information and rendering technology,image processing tools evolve increasingly powerful.Although these advanced technologies benefit for multimedia industry and can provide us good visual experience,it also makes image forgery possible.3D modeling software such as 3DMax,Maya and Softimage XSI are powerful and easy use,and they can produce remarkably realistic PRCGs that cannot be distinguished from PIM with human eyes.If those highly realistic PRCGs are used in news report,criminal investigation or insurance evidence,it will result in serious consequences for the peace of our society.Therefore,digital image forensics is becoming one of the primary tasks to identificate the authenticity,primitive,and integrity of digital images.In the thesis,the background,significance and the status of the digital image forensic technology at home and abroad are first introduced.Based on the related theory used in this thesis,the comparison and analysis of the performance of photo-response non-uniformity noise(PRNU)extraction methods in source camera identification is provided.Secondly,a method to differentiate PIM and PRCG using multifractal spectrum features of PRNU is proposed based on the difference of image acquisition pipelines.Finally,the design and development of an identification system is briefly described.The main works of the paper are as follow:Comparison and Analysis are made to the performance of the existed PRNU extraction methods in source camera identification.At present,the existed PRNU extraction methods mainly include wavelet-based wiener denoising filtering,spatially adaptive wavelet thresholding based on context modeling,and maximum likelihood estimation.Here,area under the receiver operating characteristic(ROC)curve(AUC),true positive rate(TPR)and false positive rate(FPR)are used as evaluation criterions to compare and analyze the performances of PRNU extraction methods.Experimental results and analysis show that wavelet-based wiener denoising filtering can obtain the best performance in accuracy and time complexity.An Identification scheme for PIM and PRCG are proposed using multifractal spectrum features of PRNU.Based on the principle that PIM and PRCG are come from different image acquisition pipelines,a novel identification approach is proposed by using multiracial spectrum features of PRNU.8 dimensions of multi-fractal spectrum features of PRNU are extracted to represent the subtle differences between them,and the identification is carried out by using a support vector machine(SVM)classifier.Experimental results and analysis indicate that the proposed method can achieve an average identification accuracy of 98.72%,and has good performance in stability and insensitivity to the ratios between training samples and testing samples.Besides,it is robust against some manipulations such as adding noise,JPEG compression and motion blur,and can achieve a good identification performance even the content and texture of the images are simple.An authentication system for the identification of PIM and PRCG is designed anddeveloped.It verifies and validates the effectiveness of the proposed method in this thesis.