Research On Identification Technology For Authenticity Of Digital Images

On one hand, image digitization is the product of rapid development of computermultimedia technology, and acquisition or printing equipments for digital images. It hasbeen the sign of time transiton from film images age to digital photos age. On the otherhand, the rapid development and popularization of Photoshop, ACDSee and other digitalprocessing and editing software, has made that the non-professionals can easily modify theimage to achieve the purpose of beautification, spoof or malicious tampering. There is nodoubt that beautification of digital images can bring fun and―vision feast‖to people's lives.And the direct impact of image spoof is just entertaining the public, causing no harm tosociety. However, digital tampering usually has ulterior motives, which would bring crisisof confidence to the credibility of media image, the authenticity of judicial evidence, andthe reliability of military intelligence.As an effective approach to protect content security for digital images, identificationtechnology for authentication of digital images has become research hotspots and focus inthe field of computer forensics, multimedia information security, and so on. It has importantand profound significance in news, judicial, military and political.Closely surrounding the hot topic——―identification technology for authentication ofdigital images‖, with the target of identifying authenticity for digital images and locatingtampered areas, in this study, identification for digital authenticity was researched in-depth.The identification algorithms were designed on the basis of theoretical analysis. And theperformance of the algorithms was validated through experiments. Our main researchcontents are as follows:Research on image identification technology based on digital signature: For enhancingthe security of carrying secret information by code word and increasing the robustness ofsecret information, RS codes both have a good prospect. Therefore, a novel identificationalgorithm based on RS codes digital signature was proposed in this study. Firstly, thepreprocessing is performed on the image; secondly, the preprocessed image is divided intoregions with equal size, and the RS error control coding is made to the average value ofimage blocks within the regions, which forms the digital signature; finally, the image isidentified by the digital signature. Experimental results showed that our algorithm can accurately locate the position of the tampered pixels in the image blocks, and presents goodperformance for resisting JPEG compression and increasing detection rate of the tamperedimages.Research on image identification technology based on digital watermarking: Therobustness of current identification algorithm for image sources authenticity usingwatermarking method is generally not very high, so a robust watermarking algorithm basedon DCT and SVD was presented in our study. Firstly, extract the DC coefficients of theoriginal image after DCT transform; secondly, perform SVD transformantion to the DCcoefficient matrix; finally, mofify the maximum singular value to embedd watermarkinformation. Meanwhile, in the watermark pretreatment process, the double encryptiontechnology of Arnold scrambling and Logistic chaotic sequence are used to make thewatermark information more safe and reliable. Experiments results showed that thisalgorithm can effectively increase the quantization step length when embedding thewatermark and ensuring it unsensible, which leads to enhanced robustness. Besides, theextraction process becomes very easy and convenient because it does not require theoriginal image involved.In addition, many current algorithms used to identify digital image authenticity couldnot resist casual image distortion. Focusing on it, a robust watermark algorithm based onthe image features was proposed. In this algorithm, we embed both robust and fragilewatermark to the original image, so it can not only accomplish the source identification ofimage authenticity, but also position the modified areas of image. Firstly, extract the featureareas of digital images using Hessian affine detector; secondly, embed the robust watermarkaccording to the local direction of every pixel in the feature areas; finally, embed the fragilewatermark to the areas other than the feature areas. Experiment results showed that, theproposed algorithm can resist many geometry modifications, such as rotation, cropping andso on. It can locate the modified areas accurately, which means it is able to realize theeffective detection of image tampering, while protecting the image copyright information.Research on blind identification technology of digital images: In image compositetampering, splicing is the operation which refers to copying one region of an image andpasting it into one region of another image, attempting to make an illusion. This operationusually destroys the natural statistical properties and data consistency of original images.Thus, the splicing tampering can be detected, through analyzing the statistical propertiesand data inconsistencies of images. In our study, the inconsistency in light source directionis as the basis to reveal the splicing tampering, and a blind identification algorithm based on Lambert illumination model was proposed. Firstly, the error function and the constraintfunctions are determined, where the error function is derived from the difference betweenthe actual light intensity and calculated light intensity based on Lambert illumination model,and the constraint functions is to describe the influence of different light sources; secondly,for infinite light source images, the light source direction is calculated by Hestenes-PowellMultiplier method; and for local light source images, the light source direction is calculatedby Levenberg-Marquart least square method; finally, by comparing the light sourcedirection coming from different regions of the image, it can be determined whether theimage has been tampered, and thus the authenticity of the image can be identified.Experimental results showed that our algorithm could effectively detect the splicingtampering for images with obvious lighting conditions, and compared with the originalalgorithm, it has higher correct detection rate and detection speed.In addition, copy–paste between regions of an image is another operation in the imagecomposite tampering. It often refers to copying one region of an image and pasting it intoanother region of the same image, which is to hide the important goal or make an illusion.In order to achieve the deceptive effect, counterfeiters often rotat or scale the copied regionat first, and then paste it into another region. Besides they also do some otherpost-processing to the tampered images, such as blurring, adding noise, and so on. For thecopy-transform-move-paste tampering in the image composite operation, the correspondingmodel was established in our study. And combined with SIFT algorithm in image matching,a blind identification algorithm for copy-paste tampering was proposed. It converts blockmatching into points matching in the copy-paste tampered images, based on SIFT Markedgraph feature vector. Firstly, compute the SIFT key points; secondly, in the neighborhood ofthe key point, the Marked graph is constructed depending on the maximum angle, and theMarked graph feature vector of SIFT key points is generated, which reduces the128-dimensional SIFT feature vector down to two 36-dimensional feature vectors; finally,between two key points, match their feature vectors, and connect the key points matchedsuccessfully with lines, thus the copied region and its pasted region are the two regions withmore connected lines. Experimental results showed that this algorithm can effectivelydetect and locate the tampered areas rotated or scaled and is resistant to post-processing,such as blurring, Gaussian white noise and JPEG compression, and so on. Fourthermore,compared with detection algorithm based on the traditional SIFT feature vector, ouralgorithm has higher robustness to scale transformation.In summary, our study mainly focuses on the research on image identification technology, including identification technology based on digital signature, identificationtechnology based on digital watermarking and blind identification technology. There havebeen some achiavements of our study both from theory research and algorithm applicationfrom our study, which will actively promote the development of the computer forensics andmultimedia information security technology.