The Technology Research On Digital Image Splicing Blind Detection Based On The Statistical Characteristics

With the rapid development and popularization of the Internet, computer technology and digital equipment, the digital cameras and colorful images have been widely integrated into people’s daily life. At the same time, the appearing of the corresponding digital image editing tools makes it more and more simple to deal with the modifying of digital photos and images, and it is hard to distinguish natural images from doctored images on the vision. Forging and tampering digital image becomes arbitrary, the security of digital image is facing unprecedented challenges. If false pictures are free to be used on important occasions, such as the news media, legal cases, military and other fields, it will lead the individual to produce extreme behavior, but also impact on social harmony, stability and development. Therefore, identifying the authenticity and reliability of the digital image by technology is of great practical significance and the research of the technology becomes a hotspot about digital image forensics in recent years.This thesis gives a comprehensive introduction which is about the research background, the technical classification, related research institutions and the current problems we faces. Focus on four aspects of digital image passive forensics tamper technology, forensic technology, forensic system and passive synthetic tampered image characteristics were summarized. Focused analysis of the current mainstream passive forensics algorithm is carried. A new splicing detection algorithm based on statistical feature extraction as the innovation is put forward on the basis of previous studies. The main work of this paper is as follows:(1)Research on Digital Image Mosaic passive forensics ideas, show a learning of the current different evidence algorithms, give a detailed introduction about classification algorithms, models and feature amount used in passive forensics.(2)Sum up the existing theoretical models of digital image passive forensics.Conclude a theoretical passive forensics model based on statistical characteristics on the basis of previous different algorithms.(3)Because the defect of feature extraction and recognition rate high dimensions about traditional image mosaic detection methods, this thesis carries out studies and research on the relationship between feature dimensions and splice detection accuracy about the image feature extraction.Based on the thought that image is made up by many pixels and image manipulation must cause image pixels change, this thesis puts forward to the modified characteristic dimension thought based on the pixels. And the thesis puts the digital image information entropy and the difference of characteristics of digital image fusion splicing detection algorithm. In this method, firstly extract the information entropy and differential incentives from the original image; secondly make characteristics of the fusion by using different histogram; again establish classifier model by the v-SVM(v-Support Vector Machine); at last determine whether the image is tempered. After the simulation with experimental platform, it proves that algorithm proposed in this thesis in dimensions and characteristics of the detection accuracy in Colombia stitching image library all reflect the advantages. And the algorithm provides the basis in order to achieve fast splicing detection.