Passive Image Forensics Alogrithm Based On Block Matching

With the rapid development of multimedia technology, digital image which is one of the multimedia information carriers has been widely used for it’s good properties such as scene and intuition. However, the authenticity and integrity of digital image which used in some special occasions should be protected, since it can be easily modified. The research content of this thesis is the block-matching based image passive forensics technology.The thesis firstly summarizes the research background of digital image forensic technology, and describes the importance of detecting forgeries of adding target and moving one in an image. This thesis deeply studies the forensics algorithms based on block matching from two modifications including copy-past and image inpainting. It also summarizes the advantages and disadvantages of the existing algorithms.For the copy-past tampering of adding target, there are two kinds of high false detection problems in the existing forensic algorithms. One is the self-similar real area, the other is copy source region belonging to the real content. To distinguish the source region and tampered region reasonably, this thesis reselects the binary template and calculates indexes to objectively evaluate the comprehensive performance of algorithms. A passive image forensics algorithm with reducing the probability of false rejection is proposed in this thesis. By effectively removing false detection in real smooth area and accurately distinguishing the source region and tampered one, the probability of false rejection detection is effectively reduced in the proposed algorithm. However, it fails in detecting the forgery of removing targets by image inpainting, and it can not distinguishes non-postprocessing copied source region and tampered one.To detect the forgery of removing targets by image inpainting method, locate the tampered regions of non-postprocessing and reduce time complexity of the algorithm, a fast forensic algorithm which can effectively locate the tampered regions of non-postprocessing is proposed. This algorithm is based on the existing inpainting forensic algorithms based on blocking matching. Firstly, the theory of clustering is used to classify the image blocks, and then similarity of the image blocks matches in the same cluster. The operating efficiency is improved by reducing the search scope of matching blocks. Secondly, the features of2-D phase coherence of the image are used to effectively distinguish the resource region and the tampered one in copy-past forgery, which improves the comprehensive performance of the algorithm. Experimental results indicate that the algorithm can locate non-postprocessing tamper region accurately with higher detection rate. What’s more, the operation time efficiency is improved.Finally, a simulation system for the algorithms is completed, and the simulation results are analyzed.