Research For Processing History Of Digital Image Forensic

With the widely use of digital facilities, it is more and more easily for digital images to obtain and spread. Therefore, ensuring the authority and credibility of digital images has become particularly important in the information society. Compared with digital watermarking technology, image forensic technology are more practical without embedding any information previously. Digital images can be detected by digital image forensic technologies directly. However, most forensic techniques only focus on a single specific manipulation. But in reality, people usually do not change an image just by using a single operation. Therefore we need to solve the multi-operation manipulation problems.In this paper, we focus on the processing history of digital image forensic. Our research mainly solve the problems of the double manipulations on seam carving and contrast enhancement or re-sampling. And we propose corresponding forensic algorithms. Our work includes:(1) Seam carving is a content-aware resizing method which can recognize important regions of an image and scale the image without changing the important regions. We find out the un-reasonable part in insertion algorithm and improve it. Then we propose a new forensic algorithm to detect this forgery.(2) For the research of processing history of double-manipulation, we propose a forensic algorithm surrounding the method of seam carving. Under the deeply analysis of seam carving and contrast enhancement, we propose several features to detect image forgery. And with a combination of these features, we can solve the problem of detecting seam carving and contrast enhancement.(3) And for the double-manipulations of seam carving and re-sampling, we improve the Markov transaction probability algorithm, which is used for steganalysis. The newly proposed method can describe the relationship between pixels more detailed and is fit for detecting the processing history of seam carving and re-sampling. The forensic algorithm can not only detect whether the image is altered by some operations, but also find out the order of those operations.