Research On Image Identification Based On Perceptual Hashing And Healing Brush Tamper Blind Authentication

With the high speed development of multimedia technology and image achievetechnology, the digital image is springing to people's study and life, and it become theimportant evaluation standard of people seeing things, even is higher than the role of words.The digital image age has integrated into the people's life. At the same time, image editingand processing software has increased in the number, function is more rich and operation ismore convenient, such as Photoshop, Adobe Illustrator(AI), Meitu and so on. These imageedit software make the digital image easy to modify and edit, even if the non-professionalworkers also can very easily to do image beauty or malicious tampering.The beauty to digital image shock understandably the people's vision, and besmirchesto image is usually entertaining, but the malicious tamper to image is most with ulteriormotives. If these image are used for military technology or the court evidence, which willno doubt give serious impact to social and political stability. As a effective means forprotecting digital image, the digital image authenticity technology has an importantmeaning to protect the truth of the court evidence and the news media, and has become thehot topic of multimedia information security and the computer forensics field in recentyears.Digital image content authenticity identification technology is divided into initiativeidentification technology and passive identification. The initiative identification technologyincludes the digital signature and the digital watermarking. The image identificationtechnology based on digital watermarking is to make the redundant information orcopyright information as the watermarking embedded into the image, and then identify theimage's truth according to the integrity of the digital watermarking. The imageidentification technology based on digital signature extracts the data information of theoriginal image and preserve, and then compare to the extracted information from thereceived image, according to which identify the truth of image. The passive identificationtechnology is also called the digital image blind identify technology, which analyze theanomalous change of image's statistical characteristics caused by tamper, and identify theimages' authenticity using these anomalous change and images' natural statistical characteristics. The blind identify technology is the new solution idea for identifying theauthenticity of digital image content. Relative to the "initiative" identify technology, the"passive" identify technology don't need to extract or embed the images'characteristics. Soit has extensive applicable scope and more challenges.Based on the research of the digital image authenticity identification, the paperintroduces the various means of tampering image, and respectively research on the initiativeidentify technology and passive identify technology. The main work is as follows:(1) For digital image authenticity active identification technology, proposed the imageauthenticity identification algorithm based on perception hash. First, the image ispreprocessed to enhance the capability of images' anti-JPEG compression. Then the imageis divided into two regions equally, and the top five of image's block average value areextracted and mapped to the element of GF to construct the RS code. The generated checkcode is scrambled by Arnold transform as digital signature to authenticate the originalimage. Experimental results showed that the algorithm not only can detect whether thecontent of the image is tampered or not but can accurately locate the tampered position,which improves the image on the robustness of JPEG compression. Thus this algorithm caneffectively distinguish JPEG compression operation from the deliberate tampering.(2) For digital image authenticity passive blind identify technology, proposed the blindidentify algorithm of healing brush retouch tamper. The algorithm firstly searches for eachblock in image, and inserts each block into the KD tree. Thus obtain the same value or thenearest match block, and then it uses the hierarchical clustering of the position vector toeliminate the scattered matching. Finally, it counts weight of zero connectivity ofsuspicious blocks to eliminate false positives, thereby locate accurately the tampered area.Experiments show that this method can efficiently and accurately identify the use of a classof image manipulation techniques for uncompressed images and high quality compressedimages. When it is applied to the images that have higher compression levels, accurateresults are also obtained if the region that has been retouched is sufficiently large.