Improvement of Steganography Parameters in Image against Steganalysis
Mohsen Jahanshahi, Mojtaba Hosseini · 2013
According to the daily increment application of steganography, identification of attack methods to the suspicious media has become to an inevitable affair in order to secret message detection (steganalysis).Different methods of analysis try to represent the difference between natural images and steganography images, by various images of statistical evaluation. Then, based on this difference and usually by using this intelligential methods and training of the appropriate classifier system, gain the model for breakup steganography images from natural images. In this article, the relatively comprehensive system will introduced for nine video formats, namely BMP, JPG, TIF, GIF, PGM, PNG, SWF, and JPEG2000 which has done optimal analysis methods by considering different ways of steganography. The main components of system are the number of SVM classifier that are trained for each video format on a certain number of statistical features. When one secret image arrives into this analyst system, decision making about it will be done based on classifier vote in a hierarchical structure. Experiments and training have been conducted with more than 600 images on the data base and on average more than 85% correct accurate diagnosis have obtained for steganography methods and different embedding rate, and less than 10% for improperly error diagnosis. These values in compared with other methods of rival are so significant, yet the proposed system, due to universalizability is better than other methods.