Digital Image Blind Forensics Research For Image Splicing

Digital image has become an effective medium for the information society, widely used in news media, legal cases and other occasions. But the growing popularity of image editing software makes image tampering become easy, which leads to image fraud cases constantly be exposed on major online media, resulting in serious crisis of social confidence. In order to effectively curb the image forgery, passive-blind image forensics has become the focus and hot of research in the field of information security. Image splicing technique is widely used in image tampering, so blind forensics research for image splicing can be used as the basis for image tamper detection system, which has a very important academic value.This paper has analyzed the framework and type of digital image blind forensics system, and focused on analysis of the typical image splicing blind forensics technology. Then, the research of blind detection of the simple spliced image and blurred image after splicing have been to do respectively.Blind detection of simple spliced images can be regarded as two-category problem in the statistical pattern recognition, and so the lack of neural networks and advantage of support vector machine (SVM) has been analyzed. After that, the blind detection algorithm for image splicing based on feature moments and Hilbert-Huang transform (Hilbert-Huang Transform, HHT) has been proposed. Because of the small correlation between the moments features from coefficients of image discrete wavelet transform (DWT) and the characteristics of HHT, the combination can effectively improve the classification accuracy. Based on Columbia University grayscale spliced image database, the detection algorithm has done the training and testing for real images and spliced images. Experimental results show that the average classification accuracy can rate to85.8696%, more than the existing72%.After simply splicing, the spliced image always requires fuzzy polish operation to eliminate or dilute the stitching marks. For the image with blurred edges, this paper has used band-stop type of homomorphic filtering and morphological edge to achieve the blind forensics. Compared with Gaussian filter and Butterworth-type filter, the band-stop filter is more ideal to enhance image contrast, and can effectively enlarge fuzzy spliced edge. For extracting the edge of filtered image, anti-noise morphological gradient operator, compared to edge detection operator, such as Prewitt and Canny, can get better edge in sharpness, closed and consistent. Simulation results show that this algorithm can effectively locate the fuzzy spliced edge, to realize the blind detection for spliced images with post-fuzzy operation.This paper includes33figures,2tables and56references.