Digital Image Source Identification And Tamper Forensics Based On Fractal Geometry

With the wide use of digital devices and imaging software, the acquisition andmodification of digital images is becoming easy. However, if the forged digital imagesare used in formal occasions such as news report, scientific research, evidence forinsurance and court, serious implications on the authenticity of the events and thestability of society will be triggered. Hence, the research of digital image forensics issignificant.This thesis is mainly focused on the research of digital image source identificationand digital image forgery forensics. Firstly, the background, significance and thestate-of-art of digital image forensic technology are elaborated and the main researchcontents and achievements of them are reviewed. Secondly, the related theory used inthis thesis is briefly described. Finally, three novel digital image forensics methodsare proposed. The main works of this thesis include:Fractal features based on fractal dimension and lacunarity are implemented for theidentification of natural images and computer generated graphics. Firstly, an ima ge istransferred from RGB color space into HSV color space. After that,9dimensions offractal features based on fractal dimension and lacunarity are calculated from HSVcolor space. Finally, these features are used for identification based on SVM classi fier.Experimental results show that it can achieve an average identification accuracy of95.42%for computer generated graphics, and an average identification accuracy of97.52%for natural images.A novel scheme for the identification of natural images and computer generatedgraphics is proposed based on statistical and textural features. Firstly, the differencesbetween them are investigated from the view of statistics and texture, and31dimensions of feature are acquired for identification. After that, LIBSVM is used forthe classification. Experimental results show that it can achieve an identificationaccuracy of97.89%for computer generated graphics, and an identification accuracyof97.75%for natural images. The analyses also demonstrate its excellentperformance compared with some methods based on statistical features or otherfeatures along.A novel digital image forensics scheme is proposed based on noise estimation andlacunarity texture. The lacunarity texture and the image’s noise are extracted to construct the feature space, and SVM (Support Vector Machine) is used forclassification. Experimental results show that it can be used to detect both localartificial blur and global tampering operations.The three algorithms proposed in this thesis can effectively and accurately identifythe source or the authenticity of images.