Blind Steganalysis Based On SVM For Image

As a study hotspot of information security field, steganography has made a lot of progress in the few years. Steganography is the art of undetectable communication in which messages are embedded in innocuous looking objects, such as digital images. The development of steganography not only brings in a new covert means of communication but also brings about a new threat, thus steganalysis technology comes into being. Steganalysis is a counter-process of steganography, which aims to determine if a given medium (such as image) contains embedded messages. On the one hand, it can use to reveal the illegal communication with some steganographic tools; on the other hand, it can promote the security of steganography technology and promote the steganographic algorithm into practical use. 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 blind steganalysis. In this paper, we focus on the detection of hidden messages in digital images in the case of blind steganalysis.Generally speaking, blind steganalysis first extracts several good features from images which usually change along with the size of embedding payload; then constructs a feature-based classifier to train and test the features, so as to distinguish the cover images and stego images. This paper has done some useful attempts in blind steganalysis. The experimental results have proved that our steganalysis method is effective in attacking the popular steganographic schemes. The main work and contributions of this paper are as follows:1. Concluding the feature extraction method of state-of-art blind steganalytic schemes, and pointing the advantage and disadvantage of them. It gives directive function to feature extraction method of this paper.2. Analyzing the impact of the hidden information to the gray level co-occurrence matrix and extracting gray level co-occurrence matrixes as features. As the dimensions of gray level co-occurrence matrixes are too large, the forward differences are calculated towards adjacent pixels to obtain differential images for a natural image. Then the differential images are thresholded with a pre-set threshold to remove the redundant information in order to reduce the dimensions of co-occurrence.3. We research the blind steganalysis in view of image noise. According to the additive noise model of information hiding, the hidden information embedded in image is viewed as additive noise of image, and the embedding of hidden data will change the original noise. Thus using three aspects of de-noising algorithms, wavelet analysis, neighborhood prediction to analysis and extract the noise features.4. For the difficult of the texture image detection, texture classification method is introduced to blind steganalysis. We use signal processing technology of texture classification based on local linear transformation, and extract the high-frequency coefficients of local DCT as features to detect texture images.