Study Of Digital Image Forensics And Anti-Forensics

In today’s information age, the advent and popularity of computer brought revolutionary effect to the whole society. Also, the maturity and popularization of the network and multimedia technology, together with the various digital equipments, give people the unprecedented convenience of obtaining information and entertainment. It’s so easy to obtain a digital image, spreading a digital image is also more and more easy due to the rapid development and popularity of network. The arose various digital image processing software become more and more powerful, such as famous PhotoShop, GIMP and so on. By using this kind of software people can easily modify the color, shape, even the content of the digital image to meet various requirements. However, while this kind of software affords convenience to people, they also afford the possibility of tampered image which is made by those who have ulterior motives. The tampered image used on news media will trick and confuse the public; used as the court evidence will influent the judicial fairness; used as tools to frame others up, and so on. These malicious tampered images are sure to seriously affect the social order, fairness and justice. Therefore, it’s more and more important to accurately judge one digital image is authentic or a forged one. In response to this situation, digital image forensics springs up quickly and becomes hot fields these years.The digital image forensic technique can be divided into two kinds:proactive forensics and passive forensics. Positive forensic means special information should be embedded into image beforehand, the embedded information will be extracted and checked when the authenticity of the image is confirmed. The passive blind forensics means to judge the authenticity of the image without any embedded special information beforehand. The passive blind forensics plays an incresingly important role in information forensics and attracts extensive attention due to its wide applications.This thesis follows judging the authenticity of digital image; focus on several kinds of passive blind forensic and anti-forensic techeniques. The specific works are as follows:(1) Proposed a ridgelet transform based forensic method to detect copy-move forgery. The algorithm turns the image from space domain to frequency domain. First divide the image into non-overlapping blocks, then perform ridgelet transform on these blocks, turn the obtained2-D matrix into1-D vector and then cut one part from it to form invariant feature of image block. Sort all vectors by lexicographic order, and then try to find the similar image blocks by calculating the Euclidean distance between vectors. If the distance less than the threshold value set before, then get the conclusion that there are similar image blocks in image, in other words, the image is copy-move forgery image. The experiments result shows that the proposed method in this thesis can locate the copy-move area accurately, and shows up robustness under attacks such as Gaussian noise, blurring, rotation, JPEG compression. Specially, the method shows good robustness against JPEG2000compression, which is the new compression standard and sure to be popular in the future.(2) Proposed an expanded Markov feature based forensic method to detect a kind of image forgery caused by Seam-Carving, which is a popular content-aware image resizing technique. Seam-Carving can resize the digital image without changing the details too much. That means, this technique try to avoid stretching, distorting, revising main content of the image when resizing it. Seam-Carving often be used in forged image by removing some contents, for example, a specific people or object. Till now there are few algorithms have been proposed aiming at this kind of forgery. The proposed algorithm in this thesis expanded the traditional Markov feature. The expanded feature can reflect not only the correlation within DCT block, but also the correlation between the DCT blocks. This character makes the new expanded Markov feature more sensitive to the tampered image by Seam-Carving. The experiments result shows that the detect accuracy of the proposed method is better than the traditional Markov feature based method and other existing methods.(3) Proposed an improved chaos theory based anti-forensic method, aiming at forensic methods based on JPEG images. First divide the JPEG image into non-overlapping blocks and perform DCT transform on these blocks, then using the obtained DCT coefficients matrix to estimate the distribution model of DCT coefficients before JPEG compressed. Based on the model, add the chaos noise to every subband of DCT coefficients matrix. The experiments result shows that the proposed method can erase the "gully" in histogram of DCT coefficients perfectly. Moreover, it can make the existing JPEG image forensic methods fail. Especially when the compress factor less than50, it still work well meanwhile with a maximum of maintaining the visual quality of the image.