Research On JPEG Images Forensics

Due to the authenticity and reliability of digital images,they are widely applied in many fields like journalism,law and order,and insurance,etc.With the popularity of digital editing softwares,however,images can be easily manipulated leaving invisible traces.As a result,verifying the originality and authenticity of images is becoming more and more challenging.Digital image forensics is the technique that utilizes the intrinsic fingerprint of images for authenticity identification,tampering localization and the operation history estimation.Since most images need to be JPEG compressed for transmission and storage,JPEG compression traces based forensics has been attracted more and more attention.The dissertation mainly focus on JPEG images related forensics,including robust median filter detection for JPEG image,double JPEG compression detection,quantization parameters estimation and JPEG image splicing detection,localization and attribution,which are carefully studied.The main contributions are summarized as follows:(1)We propose a robust median filtering detection method for JPEG images.Median filtering can change the pixel value in the local region,which results in the change of statistical characteristics of local texture micro-features.Then,the joint histogram of the local binary patterns is proposed to quantify the occurrence statistics of microfeatures in an image.On the other hand,median filtering can cause the change of the correlation of local pixel.To measure such correlation,the correlation coe cient of the pixel di?erence matrix is designed,which can better describe how pixel value changes introduced by median filtering.Finally,the two feature sets are concatenated as the local di?erence descriptor for the detection task.Experimental results show that the proposed method outperforms existing detectors in robustness,especially for small image patches and low JPEG qualify factors images.(2)We propose two methods for double JPEG compression detection with the same quantization matrix.The JPEG coe cient is theoretically studied and we find the truncation error is the key factor of the detection task.Then,the relationship between truncation errors and the perturbation strategy is analyzed and it reveals that modifying non-zero coe cients can cause lower probability of truncation errors than modifying the zero coe cients.Unlike existing random perturbation strategy that just indiscriminately selects JPEG coe cients for modification,in the proposed method,only JPEG coe cients with values±1 are modified by a novel modification strategy.The second one is designed by leveraging the component convergence during repeated JPEG compressions.Firstly,an in-depth analysis of the pipeline in successive JPEG compressions is conducted,and it reveals that the rounding/truncation errors as well as JPEG coe cients tend to converge after multiple recompressions.Based on this fact,the backward quantization error(BQE)is defined,and we find that the ratio of non-zero BQE for single compression is larger than that for double compression.Moreover,to exploit the convergence property of JPEG coe cients,a multi-threshold strategy is designed for capturing the statistics of the number of different JPEG coe cients between two sequential compressions.Finally,the statistical features of the dual components are concatenated into a 15-D vector to detect double JPEG compression.Experimental results demonstrate the e ciency of the proposed method,which outperforms some state-of-the-art schemes.(3)We propose two methods for JPEG compression parameters estimation.By adapting a dense CNN network to the problem at hand,particular attention is paid to the choice of the loss function.It has the capability of working under very general conditions,the improved performance in terms of mean squared error and accuracy.A method is also proposed to estimate the primary quality factor of double compressed JPEG images in the presence of resizing.By theoretically analysing the Welch Power Spectral Density(PSD)of the DC coe cients histogram of the counter-resized image,we find that the most prominent peak of PSD nonlinearly maps to the quality factor in the first compression.A simple yet e cient method is proposed to estimate the primary quality factor based on the geometric fitting.Experimental results demonstrate the proposed methods provides superior performances.(4)Detection of inconsistencies of double JPEG artifacts across di?erent image regions is often used to detect local image manipulations,like image splicing,and to localize them.We move one step further,proposing an end-to-end system that,in addition to detecting and localizing spliced regions,can also distinguish regions coming from di?erent donor images.We assume that both the spliced regions and the background image have undergone a double JPEG compression,and use a local estimate of the primary quantization matrix to distinguish between spliced regions taken from di?erent sources.To do so,we cluster the image blocks according to the estimated primary quantization matrix and refine the result by means of morphological reconstruction.The proposed method can work in a wide variety of settings including aligned and nonaligned double JPEG compression,and regardless of whether the second compression is stronger or weaker than the first one.We validated the proposed approach by means of extensive experiments showing its superior performance with respect to baseline methods working in similar conditions.