Research On Steganalysis For JPEG Images

Steganography makes use of the redundance of human sense and the statistical redundance of digital medias inorder to hide the presense of communication by hiding the secret message into the digital media files which are translated on the safe and public channels.There are many kinds of digital media which can be used to hide seceret message, such as audios, videos and images, etc.With the development of the Intenet, the digital images are widely used. The JPEG images have become a good choice for the thanslation of image data on the internet because of the high compressiong ratio, good quality and small storage. So, the JPEG images can be used to hide the secret messages and don't arouse any suspicion. Thus it is very important to study on the steganalysis for JPEG images.This paper has a study on the steganalysis for JPEG images.The main works of this paper are summarized as follows:1,The history, the classification and the application of the information hiding are briefly reviewd. Then, the steganography are introduced and the characters, the evaluation criterions are dicussed. Finally, the importance of the steganalysis has been paid much more attention. The research works about the steganography and steganalysis in recent years are also summarized.2,The classical steganography algorithms for JPEG images are introduced and the problems of these algorithms are discussed. Then the steganalysis methods for those classical steganography algorithms are summarized in two categories.3,A method based on the differences of coefficient combinations of 0,1 for detecting F5 steganography are proposed. F5 is one of the classical steganography algorithms in JPEG images. The coefficients of 0,1,-1 are changed obviously after embedding the serect messages in the JPEG images using F5 algorithm, thus the shrinkage is brought about. The combinations of coefficients of 0,1,-1 have 9 situations altogether which can be called the coefficients combinations of 0,1. We find that the coefficients combinations of 0,1 at the two neighboring frequency places following the Zig-zag scan in the 8×8 blocks of the JPEG images change differently after embedding the secret message. Then we take 12 characters based on the differences of coefficient combinations of 0,1 in the 8×8 block and between the two horiziontal and vertical neighboring blocks. Thus, the Support Vector Machine (SVM) is used to classify the cover and the stego images. The experiment results demonstrate that the proposed scheme is effective for detecting F5 steganography.4,An algorithm based on the difference of absolute value of coefficient differences for detecting the double compresstion in JPEG images are proposed. The standard codes of steganographic algorithms F5 and OutGuess always decompress the cover JPEG images into the spatial domain and compress it again before embedding, usually use the default quality factor(F5 is 80 and Outguess is 75). If the quality factor used in recompression differs from the JPEG image's primary quality factor, the detection accuracy of some steganalysis algorithms are negatively influenced. We compare the histograms of single-compressed and double-compressed images and find that both the absolute differences of DC and AC coefficients, the absolute differences of AC and the next AC coefficients in the 9 low frequencys in the 8×8 blocks of the JPEG images change obviously after double compression. Then, the Support Vector Machines (SVM) is used to detect an image whether it is double compressed or not. The experiments results show that the proposed method in this paper can practise a good performance while the fales positive rate can be control under 5%.In the end of this paper, the research works are summarized and the future work directions are discussd.