Invisible In The Image Information Detection

Based on the statistical analysis of cover-image & stego-image,utilizing the classifier of Support Vector Machine, two kinds of steganalysis algorithms with high detecting efficiency are proposed. To estimate the amount of hidden information, a novel algorithm is also designed for the intending information extraction. The research includes the following sections:1. A steganalysis algorithm for the detecting of 256 gray-scales bitmap was designed. A variant X was defined as the statistical results of the differences of neighboring image pixels. The zero-crossing count, the maximum, the positive & negative count, and the count of peak points of X were used as characteristic vectors of a 2-class Support Vector Machine classifier to identify the cover-image and the stego-image. The algorithm is effective to most of the steganograhpic algorithm. When it was applied to the detection of cover-images and different kinds of stego-images, the average detecting accuracy to the cover-image and the stego-images are 94.9% and 91.3%,respectively. 2. Based on the principle that the DCT coefficients of JPEG image will be changed after embedding information into the cover-image by Jsteg algorothm, a steganalysis algorithm to detect stego JPEG files built by Jsteg algorithm is designed. Using the frequency of occurrences of the DCT coefficients as a feature vector, the Fisher Linear Classifier is adopted to classify the cover-images and the stego-images. However, due to the worse classifying performance of the Fisher Linear Classifier to the stego-images with little information embedded, the Kernel Fisher Discriminant is used to improve the detecting accuracy. After introducing the method of Kernel Fisher Discriminant, the detecting accuracy to the JPEG image with 20% embedded data was increased from 59% to 63%. 3. Based on the truth that the fluctuating level of the emerging frequency of DCT coefficients increases with the content of data embedded in the JPEG image, the multi-class Support Vector Machine is utilized to estimate the rate of embedded data. The average detecting accuracy of this method to the stego-images with different embedded-data rates reaches 89%. 4. An effective method is designed to prove the...