Research On Blind JPEG Image Forensics And Related Problems

With the development of digital technology, image editing tools and image acquisition equipments are widely used and updated, even people with little professional knowledge can tamper a digital image easily, and this tampered image is hard to be distinguished by naked eyes, if those tampered images are abused in the area of judicial, news, etc, it may cause a series of problems both in social development and political stability. As JPEG is the most commonly used storage format of digital image, the study on the technology of JPEG forgeries appears quite important. The main contribution of this article as follows:(1)A blind image forensics based on block compression effects is proposed. In the process of JPEG compression, the image is divided into not-overlapped 8×8 size blocks, and those blocks are quantized with its corresponding quantization coefficient of the quantization table, through the study, the difference between the high frequency coefficient in the quantization table and the low frequency coefficient in the quantization table is large, therefore, those image blocks will appear a characteristic in the process of quantification, and this characteristic is mainly appeared in the image with JPEG compression history. In this paper, we using this characteristic to generate a blind image forensics based on block compression effects, and this method can identify an image is a forged image or not without any priory information.(2) A method of image forgery localization in JPEG images is proposed. A tampered image is composed of forged areas and background areas, first, through the way of establishing and constantly improving the quantization table database, and then combine the existing method of estimating the primary quantization table, match the estimated table with the quantization table in database, we can get the correct quantization table, which is the background areas’priory quantization table. Through the research, when compressing the tampered image with the background areas’ priory quantization table, we can find that the background areas will generate a phenomenon, we call the phenomenon "JPEG Ghosts", and then using this phenomenon to propose a method of image forgery localization in JPEG images. This method can automatically localize the tampered area and is effective under post-processing operations such as rotation, resizing, blurring, etc.