Research On Passive Blind Forensics Of JPEG Forgery Image

With the popularity of digital cameras and the rapid development of networktechnologies, digital images and image processing software have been widely used andspread in many areas. At present, the tampered image events which emerge in an endlessstream have affected people’s normal life to some extent. Besides, the majority of doctoredimages are always saved in JPEG format. Therefore, this paper investigates the JPEGforgery images, and proposes the blind detection methods. The main contribution of thisthesis includes:(1) A method of differential edge detection is proposed to detect JPEG forgery image,which is based on the analysis of gray value change on the splicing edge. After calculatingdifferent directions of the gradient in three color components, we find the outliersaccording to the corresponding element of direction changes in the local area ratio. Then athreshold-based method is employed to classify the natural edge and the splicing edgepixels. At last, we superpose the splicing edge pixels and export the denoising results. Theexperimental results show that the algorithm can detect more completely the manipulatedtrace of the forgery images, which experienced different JPEG compression history.(2) Aimed at the problem that the tampered images have experienced JPEGcompression once, an efficient algorithm is proposed based on estimating the quantizationtable. Firstly, the quantization table of the spliced portion is estimated by the histogram ofthe DCT coefficients. Then the doctored area is detected through the inconsistenciesbetween the tampered area and the background region. At last a series of experiments alsohave shown the accuracy and effectiveness of the algorithm in actual detection oftampering images. The experimental results show that the algorithm can not only estimatethe quantization step in a relatively short period of time, but also accurately locate thetampered area this kind of doctored images.(3) To locate the doctored images that have experienced JPEG doubly compressed,we firstly analyse the features of the DCT coefficient histogram which quantized by thesecond quantization step, and then estimate primary quantization matrix in background according to the DCT coefficient histogram. Combining with the estimated primaryquantization matrix, the tampered image region was located further more by using theJPEG compression characteristics from compression history in background and tamperingareas. The experimental results show that, this algorithm can effectively detect the tamperlocation and shape of the tampered images. Besides, it also has good detection effect totest the forgery images that experienced the rotation, scaling and other post-processingtampering means.