Write Analytical Methods, Based On An Implicit Pollution Data Analysis

After studying a class of steganographic algorithm based on the spatial Least Significant Bit(LSB) replacing in the still images, a steganalytic technique based on contaminated data analysis for stego-only is proposed in this thesis by using the knowledge and methods of probability, statistics and image processing. The method can not only fast and reliably detect the existence of hiden message embedded in images, but also accurately estimate the amount of hiden message embedded by sequentially or randomly scattered algorithm.For the needs of copyright protection and secrecy communication, the people have more increasing demands for information security, which makes the information hiding technique be a new important research area in the field of network information security. Because the government restrict the intensity of demotic encryption algorithm, and the encryption uses the randomicity of sequences to confront the attacker which is an obvious limitation to easily draw others' attention. So the people do not absolutely rely the communication security on encryption but turn to the research and application of steganography. Hence how to search and detect the secret messages transferred in network is crucial and practical to prevent crimes and safeguard the national security.In this thesis, based on the thought of contamination model, we proposed a new steganalysis method that founded on analysing some kinds of attack algorithms against steganography using spatial LSBs in still images as carriers. The method regards the replaced LSBs as contaminated points, and separates the pixels whose value are odd into two parts, then looks the processed noise data upon the samples come from a contamination distribution. Due to the means of four group of noise data having changed regularly after the secret message embedded in the image, we built the statistic r to quantify the dominance and confirm the threshold value. Consequently we can fast and effectively detect whether there are secret messages or not in the images. Whatever the messages were embedded sequentially or scattered randomly, the algorithm we proposed is effective to attack this class of steganography methods.Then we detailedly analysed the algorithm in theory and found that the function relationbetween value R and the embedding ratio of secret message q is q =1/(1+R). Therefore, thismethod can not only effectively detect the stego-images, but also accurately estimate the length of the embedded secret messages. We also analysed the error existing in the estimation of ratioand gave the material correction method.In the last part of thesis, after comparing the algorithm we proposed with the classic RS and PS steganalysis algorithms, we found that its whole capability has improved and excelled two of RS and PS algorithms in some aspects.