Digital Image Source Forensics Using Statistical Model

The digital image is easy to edit and easy to forge,which makes the authenticity of multime-dia in the digital age extremely challenging.For this reason,digital image forensics technology focusing on digital image acquisition equipment,content authenticity and information integrity has become a hotspot in multimedia information security research.However,the existing digital image source forensics algorithms have also emerged many problems,such as the large training set required by the algorithm,the high feature dimension and the lack of theoretical proof.In response to the above problems,this dissertation examines the following issues.First,a source forensics algorithm for natural images and computer generated images in non-RAW format is proposed.Because the imaging mechanisms of natural images and computer-generated graphics are different,there are large differences in statistical characteristics between the two types of images.In order to effectively extract the statistical features of the image and reduce the interference of factors such as image texture,the residual noise of the image is ex-tracted step by step through the two filters designed.On this basis,the Gaussian distribution model is established for the extracted residual noise by means of hypothesis testing theory.The residual noise of different images obeys the Gaussian distribution model with different param-eters.The experimental results show that under the condition of false alarm rate of 1.2%,the positive detection rate of the proposed algorithm for natural images is 95.33%,and the positive detection rate for computer generated images is 96.44%.In addition,the theoretical upper limit of the forensic algorithm and the positive detection rate are analyzed,and the forensic algorithm has better robustness under image post-processing attack.Secondly,the variance values of residual noise in RAW format images are studied for post-processing operations such as CFA interpolation,white balance and gamma correction.The post-processing operation is simulated using a bilinear interpolation algorithm,a gray world algorithm,and a gamma correction algorithm.Source forensics of natural and computer-generated images of mosaics,CFA interpolation,white balance and gamma correction using the noise distribution model mentioned above.The experimental results show that under the condition that the false alarm rate is close to 1%,the positive detection rates of natural images are 54%,97.1%,95.2%and 72.4%respectively.CFA interpolation introduces a high correlation between image pixels,while white balance and gamma correction attenuate this correlation.In summary,the digital image source forensics algorithm based on statistical distribution model not only overcomes the shortcomings of the current forensic algorithm,but also maintains a high positive deteetion rate.The recognition accuracy of the algorithm is theoretically proved.Different from the current research on image source forensics algorithms for non-RAW format images,this work also reveals the inherent characteristics of RAW format images introduced by post-processing,as well as the typical relationship between the intrinsic properties and the accuracy of image source forensics algorithms.