Information Hiding System Based On JPEG Images

With the rapid development of information technology and the wide use of the internet, as anew cryptographic technique, information hiding becomes a cutting-edge of the informationsecurity area. As two important branches of information hiding, steganography aims at sendingsecret message covertly by embedding it into public digital multimedia, and steganalysis studieson how to detect, extract, and destroy the hidden messages.Targeting on development and application requirements of the information hidingtechnology, some researches are done on digital image based blind steganalysis in the presentthesis. The main bodies of the studies would be outlined as follows:Proposing a blind steganalysis method based on a special condition.Proposing a searchingscheme for cover image. The attackers can decide the webs by way of the watermarking in thesuspicious image after getting it through Security Division et al, and download the images withthe same size; then they can be transformed to spatial format, and the suspicious cover can bechosen through rough sieving, and the final cover image can be chosen through fine sieving afterbeing transformed into the same format as the stego image.Proposing a decision scheme forcommon steganographic methods after getting the cover image. The decision can be made andthe steganographic scheme or software can be decided with the help of embedding scheme ofcommon steganographic software.The experimental results demonstrate that the proposedscheme can search the cover and recognize the steganographic method exactly.Proposing a blindsteganalysis method based on SVD and low-rank matrix recovery.Given the number of coverimages being larger than that of the stego images, we talk about the effect of different ratios onblind steganalysis methods. The experiment dictates that the best classification can be got as thenumber of the two kinds of examples is the same, and it obeys the fact that the best result can begot when obtaining samples of the same number under the condition of not enough trainingsamples. Differential image is computed, then SVD is made for the differential image, and thesmaller singular value vectors are chosen as a part of the feature; Low-Rank Matrix Recovery ismade on the image and the noise is got, then SVD is made for the noise, also choosing thesmaller singular value vectors as another part of the feature. The proposed scheme has anadvantage based on the experiment using the steganographic methods of replacing、matching andStegHide.