Research On Blind Identification Algorithm Of Image Splicing Based On Statistical Model

The rapid development of computer science and technology, the rapid spread of digitalcameras and a variety of mature image editing software flooding into the entire Internet, makeit easier for people to take pictures, and then be exposed to more digital images. Thus, digitalimage has go deep into the daily life, especially when image-processing software todayenhance usability to make modifying images and landscaping photos become the growingtrend of digital information age. At the same time, with the advent of the age of socialnetwork, people are more willing to share their experience, thoughts, feelings, digital imagewhich is most representative. People enjoy sharing photos, sharing knowledge, but at thesame time, some people deliberately tamper photos and publish to the Web, tamper politicalphoto to confuse people or landscape photograph of a product online for the purpose of profit,which affect people’s normal life. Then, the "see with one’s own eye" have begun to bequestioned. Thus, digital image forensics has gradually become noticed topic, and should betaken seriously.Forensics of digital image or prints is divided into three areas: the forensics ofauthenticity of digital image sources, forensics of the authenticity of the digital image contentand forensics of restoration of blurred and degraded image. the forensics of authenticity ofdigital image sources is mainly about getting the device and its corresponding imagingprocess for those different images. The forensics of authenticity of digital image sourcesrefers to being informed the imaging equipment and imaging process under the premise of anunknown image source. Forensics of the authenticity of the digital image content refers to thedetection of image tampering without prior identification information embedded andrestoration of the tampered image if possible. In the process of image formation, transmissionand recording, influenced by many factors, the image quality will decline, such as blurring,distortion and noising, which are called fuzzy and degraded of image. Therefore, therestoration of degraded and images are needed. Due to the different ways to get images, blurdegenerate form is varied. If the type of degradation mechanisms and processes are very clear,we can use the anti-process to fulfill the restoration of degraded image. The quality of imagerestoration depends on the degree of precision of image degradation process prior knowledge.This paper describes tampering models of digital image and focuses on image splicing whichis the most common tampering operation. Blind identification algorithm of image splicing isintroduced in detail. The various features for blind identification algorithm of image splicingis summed up. After discussion of these characteristics, the advantages and disadvantages of these characteristics are analyzed. Then an improved algorithm has been proposed.Image splicing is one of the most common means of digital image tampering operations,which put two or more images together into a fake picture without further treatment, such asedge smoothing. Image splicing will destroy the statistical properties of natural images, whichmeans there will be inconsistencies and discontinuities between pixels of spliced image whichcan be captured by statistical model features.In order to improve the detection accuracy of spliced images, a new blind detectionbased on fuzzy run-length was proposed in this study. Firstly, edge digital image gradientmatrix is calculated to obtain the gradient direction of each pixel which is then quantified toone of vertical, horizontal, main diagonal, and minor diagonal directions; Secondly, accordingto the gradient direction of each pixel, histogram is calculated, including fuzzy run-lengthhistogram, fluctuation degree histogram and fluctuation count histogram. Thirdly, extract thethree lowest order moments of characteristic of histograms as features for splicing detection.Finally, train and classify the above features using SVM, by which the spliced images can beidentified from the natural ones. The experimental results showed that, when testing on theColumbia image splicing detection dataset, the detection accuracy of the proposed method hasimproved a lot, which has a good growth prospect.In order to enhance the sensitivity of the feature vector of the splicing tampering, a newblind detection based on the Visual Attention Model （VAM） was proposed in this study.Firstly, the edge conspicuous map （ECM） can be created by an improved OSF-basedfiltering approach, then extract fixations by VAM, and locate those on boundaries byconspicuous edge positioning method, accordingly the key feature fragments can be captured;Secondly, extract Extended Hidden Markov Model （E-HMM） features from each waveletreconstructed images of Cr channel of the fragments, and reduce their dimension bySVM-RFE; Finally, train and classify the above features using SVM, by which the splicedimages can be identified from the natural ones. The experimental results showed that, whentesting on the Columbia image splicing detection dataset, the detection accuracy of theproposed method was higher and the computation cost was lower than other method.