Forensics Of Image Operator Chains Including Blurring Operation

The development of digital image processing technology promotes the popularization of image retouching and tampering. With the increasingly serious image trusting problem, the research of digital image forensics has been paid more and more attention. Digital image forensics is a research field of analyzing the statistic features of the content of digital images with the aim of identifying the authenticity, integrity and primitivism of testing objects. The research of image operation chains is a new branch of digital image forensics, and will be an important part in future studies.Operation chain is defined as a set of manipulations that processed in sequence in the operation history of an image. In this article, we introduced the development of research in digital image forensics, analyzed the practical application of blur operation in real life, and proposed two research topics in forensics of binary operation chain that contains blurring. Novel features and detection algorithm are proposed for each topic, and abundant image databases are built for testing the accuracy of proposed method.A brief introduction of our work is as follow:(1) Detection algorithm of binary operation chain that consists of blurring and sharpening is proposed based on the NSCT (No-subsampled Contourlet Transform) domain. The high frequency information of an image will weaken or enhance after blur or sharpen operation applied. NSCT decomposes an image into frequency information at different scales, so that the distribution change of decomposed coefficients in NSCT domain reveals the relating tamped traces. In this paper, three sets of features in NSCT domain are proposed to detect processed trace of the operation chain, which are edge type features, extremum features and direction similarity features. The experiment results show, the operation identification accuracy of proposed detection method is satisfied.(2) Effective algorithm is proposed to detect binary operation chain that constituted of blur and noise addition. The fluent transition between adjacent pixels is damaged by noise addition to an image. The alteration also leads to isolated obvious change of pixel value in neighborhood, especially in those flat areas. In this article, statistical features of differences among neighbour pixels are proposed to capture the trace. The direction similarity feature in NSCT domain is modified to detect binary operation chain of blur and noise addition. The experiment results show, with the use of SVM, algorithm can perform well enough in classifying different kinds of tampered images.