Research On Blind Digital Image Forensics For Authenticity Detection

Since digital image is easy to copy, edit, and deliver through the Internet, it has become important kinds of information carrier, which has been shown in every field of our life, including news media, science research, and politics and laws. However, its development also makes it easy to tamper an image, while illegal tampatation on digital images always leads to bad influence to the life of people. As a result, the way to identify whether an image has been tampered has become an urgent problem to be solved.Digital image forensics means to use some kinds of technology to identify the digital image source, and check whether the image has been tampered. Digital image forensics is composed of two categories, the active forensic technology and blind forensic technology. Among which, active forensic technology mainly derives from digital watermarking technology. The main idea is to preprocess images, add watermark to images. Then we can check whether the image is tampered by checking the watermark in the digital images. However, blind image forensic do not need to preprocess image by digital watermark or digital signature, while implement image authentication by directly analyze the image itself. The theory of blind digital image forensic is that the intrinsic statistic property is inevitably changed after tampertation. Since blind image forensic do not need to preprocess images, it has good application perspective.For the existing problems in the field of digital image blind forensic, we first summarize the related works and theory for blind forensics, then we discuss and analyze the limitation of existing blind forensic method. Existing methods mainly forcus on tackling image completion authentication and image source authentication, but they do not tackle the more attracting problem of image content authentication. We propose the importance of Blind Digital Image Forensics for Authenticity Dectection, and choose clone detection and splicing detextion for further research.For clone detection, we propose a method based on multi-resolution historgram. It use robustness and distinguishness of multi-resolution historgram to extract224dimention feature vector for image blocks, which constructs feature matrix. Then match feature vectors to check clone opertation.For splicing detection, we convert the problem of identifying whether an image is spliced to a problem of classification. Through analyzing the property of splicing operation, we propose a splicing detection technology based on Support Vector Machine. Experiment shows that our method is effective.Clone operation and splicing operation can be considered as generating copy of a part of the orginal images, so we consider tackling operation detection problem by local copy detection, and propose blind forensic based on local copy detection. First, we extract feature vector through feature fusion, then we use machine learning based hashing to map feature vector into compact hash code, improving detection efficiency.In our research, we propose several methods which successfully solve the problem of Blind Digital Image Forensics for Authenticity Dectection. However, in terms of further research, there are many problems need to do exploration and perfection. We consider to improve the performance of Blind Digital Image Forensics for Authenticity Dectection in many aspects like exploring semantic of images, user feedback information understanding for image authentication, and so on.