Research On Passive Forensics Methods Of Digital Image Region Duplication Forgery

In today’s digital era with the rapid spread of information, digital images have profoundly affected our daily study, work and life. The phenomenal development of mobile internet makes this trend reach unprecedented heights. However, with the popularity of various powerful and easy-to-learn image editing softwares, people hardly need special expertise to freely modify a digital image with great ease, and forgery images are difficult to distinguish by naked eyes. Digital images with malicious attacks have penetrated into all aspects of real life, and breaking the traditional idea of "seeing is believing" is directing trust crisis in today’s society. Faked images would make the severely negative influence on various aspects of society, such as new reports, business promotion, academic research and court evidence-providing, thus the study on methods of image forgery forensics is of great practical significance and application value.Aiming at region duplication image forgery, the dissertation carried out mainly centered on the main line to identify the authenticity of digital images using the methods of passive forensics. Further researches were conducted on the detection algorithms of region duplication forgery which were resistant to the attacks of intermediate processing operations and post-processing operations. In view of the present popular JPEG image, we modeled different situations of copy-paste forgery with JPEG images, and studied a method to identify the authenticity of digital images by means of utilizing block artifacts grids produced in the process of tampering.The main innovations and contributions of this dissertation are as follows:1. Region duplication forgery detection methods based on block feature matching were studied. Aiming at the shortcoming that most of the existing detection algorithms are not very robust to post-processing attacks, combining the robustness of quantized DCT coefficients with the stability of the largest singular value, a robust passive forensics method based on block feature matching was proposed. A series of experiment demonstrate that the proposed approach is not only robust to multiple region duplication forgery, but also resistant to blurring, AWGN, JPEG compression and their mixed operations.2. Region duplication forgery detection methods based on local points detector and feature matching were studied. Distribution characteristics of dense Harris feature points were analyzed, and a robust passive forensics method based on local points detector and feature matching was proposed, which made use of the combination of a variety of local binary pattern operators for feature extraction. The proposed method enhances the robustness to rotation and flipping attacks of copied areas, and solves the problem of image forgery detection for smooth areas and small areas.3. In order to address the problem that the existing detection algorithms can only identify pair-wise similar regions but cannot locate the forgery region accurately, image average information losses were analyzed under different compression historical circumstances, and in view of HSI color features of SIFT points, a novel method for automatic detection and localization of the tampered region was proposed based on estimating the offset of double JPEG image compression, which could effectively locate the forgery region in different copy-paste forgery situations with JPEG images.4. Based on the model of copy-paste forgery with JPEG images, different response characteristics of JPEG coefficients change rate for double compression operation were deeply analyzed, which resulted from the inconsistency of block artifacts grids for original regions and forgery regions, and a robust passive detection method with the single process was put forward to realize integrated detection and localization of copy-paste forgery in one image and composite forgery between two images.Finally, a comprehensive summary of this dissertation was made. The remaining problems were discussed, and novel research directions were pointed out for the future works.