Steganalysis Research Based On JPEG Images

Steganography and steganalysis are important issues in the field of information security during the era of Internet. They are drawn more and more attention from governments, military, security agents and research institutes all over the world. The purpose of steganalysis is to detect the existence of secret message, estimate the length of the secret message and even extract the message. Now most of the studies are special steganalysis, but universal steganalysis is relatively fewer, and the steganographic tools in the real life are often unkonwn, so universal steganalysis is the inevitable trend in development of steganalysis . Among many image formats used in Internet, JPEG image is the most popular. This thesis analyze existing steganography methods based on JPEG and do some studies in the statistical variance of cover image as well as the design of the classifier. The main research work and contributions are as follows:1. Analyze popular steganographic methods based on JPEG and summarize their influence on the statistics of the cover image.2. Propose two new multi-feature universal steganalysis methods. The two methods both combine with the feature of the moments in the frequency domain of wavelet sub-band's histogram and block effect. The first method uses inter-block dependency as its block effect characteristics, while the second method uses the countdown of the block characteristics after high-pass filtering as its block effect charateristics. Experimental results show that both of the two methods can have a good performance, which have obvious advantages compared with the previous universal steganalysis methods. Although the second method only increase one-dimensional characteristics relatively to the moments in the frequency domain of wavelet sub-band's histogram, but the classification effect it can achieve is better than that brought by the way of increasing multi-dimensional inter-block dependency. Moreover, it has the advantage in complexity and speed.3. Apply high-pass filter into steganalysis at the first time, prove—theoretically and experimentally—the effectiveness of the block characteristics after high-pass filtering. The research work points out a promising way to general image steganalysis.4. Convert steganalysis problem into pattern recognition problem to solve, choose features according to the standard of"good"feature defined in pattern recognition "polymerization in the categories, scattered among the categories". By comparing the two methods, we find that the key of extracting features is not the number of dimensions, but whether the feature is"good".