Research On Digital Image Forensics Of Processing History

To identify the image's authenticity,digital image forensics technology emerges.The research content involves identifying the source of image,distinguishing the image captured from the real scene and that generated by computer,detecting the image oper-ation applied in the image,and revealing the action of tampering the image.Although,the digital image forensics technology has got some great achievements,the existing tech-nologies consider only easy forensics problem in the ideal scenario.That makes proposed methods can not be applied widely.This thesis further studies detecting multiple oper-ations(so called the operation chain),which is close to the real forensics scenario.By means of analysing the variation of statistics characteristics of images due to the opera-tion chain,we construct the model of the operation chain.The feature can be explored and be used to design the detector.It is significant to reveal more details of the processing history of the image.Our main contributions include:(1)By analysing the variation of the single-operation feature in the case of operation chain,the robustness measurement of the single-operation feature is proposed.Two robust single-operation features can be used to form the operation chain component fingerprint.One robust and one fragile single-operation feature can be used to form the operation chain order fingerprint.The detection includes two steps.The first step is detecting the component of the operation chain.The second step is discriminating the order of the operations.(2)It is the first time to raise the forensics problem of the operation chain''histogram equalization and dither-like operation" in the forensics research.We analyse whether histogram equalization can produce a uniform distributed histogram,which is a controversial problem in the academic community.By theoretically modeling and analysing,we finally infer that the uniform distributed histogram is the fingerprint feature of this operation chain.In order to distinguish the specific dither-like oper-ation,the local binary pattern feature is introduced.In addition,the proposed algo-rithm can be used for image tamper detection.(3)It is the first time to raise the forensics problem of the operation chain:"JPEG-Resampling-JPEG".In the pixel domain,we exploit the transformed block artifact grid(TBAG)feature,which can be used to determine if an image was undergone the operation chain,even can also be used to estimate the resampling factor.By implementing the scaling operation on the DCT domain,we find the relationship between DCT coefficients and the resampling factor.Then the DCT domain feature detector is designed.(4)The mathematical model is built for the activity of forensics and anti-forensics oper-ation.The theory proves that the optimal anti-forensics operation can be resisted by the detector using random feature selection strategy.The obtained theoretical result is applied to detect the operation chain consisting of two operations.The first one is the adaptive histogram equalization or median filtering.The second one is an feature domain or an pixel domain anti-forensics operation.The experimental results verify that the proposed method leads to only a small decrease in detection performance in the absence of anti-forensics operation,and can increase the performance in the presence of anti-forensics operations.