The Research Of Digital Image Copy-move Forgery Blind Detection Based On SIFT

With the coming of the big data era and the development of modern networktechnology, images are becoming more and more popular in our society and on theInternet, which bring unprecedented convenience to people’s communication. However,due to the appearance of faking and tampering images, whether a image is real ortampered or not is becoming increasingly unreliable. The behavior of tampering imageswill bring disastrous consequences. Therefore, the detection technology for theauthenticity of the image has become a hot spot in modern society.In this paper, the problem of blind detection of digital images copy-movetampering is investigated, First, it introduces its basic concept and several importantmathematical models. Then, it analysis the characters and the process of the algorithmof based on SIFT image blind detection, and does some improvement on it.In order to achieve the effect of visual confusion, forgers usually use the geometrictransformation or add some necessary subsequent processing operations. Existing SIFTfeature extraction algorithm cannot deal with these cases perfectly. This paper firstlyextracts the SIFT keypoints, and then calculate the texture features of keypoints aroundthe center of the neighborhood. The both features are described as the features of akeypoint.To the problem of big error of calculating the transformation parameters oftraditional RANSAC algorithm, this paper first randomly selects some non-collinearpoints, roughly calculates the initial transformation matrix, calculates the remainingpoints distance errors according to the initial transformation matrix, finally uses thedual-threshold method to examine data and get the final transformation parameters.To confirm the feasibility and ability of the algorithm, use the Matlab software todo some simulation experiments on the copy-move images. Give the test results ofdifferent means of tamperings and image detection accuracy and pixel detectionaccuracy. Experimental results show that the algorithm is effective and has a stronganti-noise and anti-JPEG compression ability.