Research On Universal Steganalysis Technology Of Digital Image

Steganography and steganalysis, as the key technology is information security field, has attracted many researches and practitioners to engage on it and made large amount of progress. The purpose of steganalysis is to detect hidden messages in media. Also, it puts forward the dubiety about hidden messages in media. The state-of-the-art steganalytic schemes can be divided into two categories. One is called specific steganalysis; the other category gets its name by universal steganalysis. This paper mainly focuses on the general detection of hidden messages in digital images.On the research of the universal detection in hidden messages, the NRCS and the color image in the benchmark image databases of Washington University are chosen to be the experimental image, the noise characteristics and texture feature of the image can be distilled as mathematical statistical feature, Support Vector Machine has been employed to classify the image characters distilled, then, the original image and the hidden image can be distinguished.From the angle of image noise, this paper made a research about the universal detection method of image steganography. According to the additional noise model, there will be some change in intrinsic noise characters of image. Based on the characteristics of noise variation, a method of multi-noise characteristics based on digital image steganalysis was proposed. Firstly, this method extracts the three image noise feature from wavelet analysis, image denoising and neighbor prediction. Secondly, it can calibrate the noise characteristics to make the noise characteristics reflect the changes of embedded watermark better. Finally, in order to judge the image whether contains hidden information, nuclear function RBF was selected as the support vector machines inner product function to classify. Through testing the four typical steganographic methods of the LSB, Cox's SS, F5 and JPhide, the results show that this method can achieve blind detection analyses of hidden information effectively.In addition, contacting the image texture classification and steganalysis problems, proposing the imbedded hidden information can be seen as introducing the more random texture viewpoints. So, this paper, combining the texture classification method, made a regional linear transformation for the color image, extracted universal steganalysis features from the normalized histograms of the factors in the regional linear transformation, and classifying the carrier image and steganography image using support vector machine classifier. The detection result of four classical steganalysis methods (LSB, JSteg, OutGuess and F5) show that: the proposed method has a good effect in LSB steganalysis method, and the detection rate seems rather high when the embedding rate is high.