Research On Image Steganalysis For LSB Matching

With the status of information security becoming more and more important, research on information hiding is taken by many research institutions all over the world. Steganography and digital watermarking are two main sub-disciplines of information hiding. Steganalysis, as the opposite technology against steganography, gets researchers'more attentions. Secret messages are often embedded into the digital multimedia, such as image,audio and video, and transmitted in an unnoticed way through the public channel(especially the Internet). The purpose of steganalysis is to detect, extract, restore and destroy these secret messages.LSB steganography in spatial domain is widely used, because its simple algorithm and large amount can be hidden. LSB replacement is one of them, which is advanced long before and have a wide application so far. It replaces the least significant bit of a cover image by secret messages in some way, and forms a stego image with other seven bits. LSB matching is proposed as an improvement of LSB replacement and resolves the problem:the sample pairs of image histogram, so its steganalysis is more difficulty. At the same time, LSB matching steganalysis in condition of low embedding rate and uncompressed high-resolution raw images are difficulties and hot points of it.This thesis works on steganalysis techniques in LSB matching based on the in-depth analysis of steganography systems in spatial domain. The main work and contributions of this thesis are summarized as follows:1. Modeling the embedding of secret messages in LSB matching as a corruption of additive noise to the "pure" cover image, so image restoration is introduced to reconstruct the cover image. Because the embedded secret messages in LSB matching have the same characteristics as impulse noise which can be filtered by median filter efficiently, an improved method based on median filter is presented. Also, through the experiment, the necessity of filtering restoration is testified. And, an efficiency approach is proposed to resolve the influence of different image contents on secret messages detecting. By comparing the testing image and the estimation of cover image, the number percentage of value 0 in HH subband wavelet coefficients is introduced as a difference feature to identify stego image.2. Aiming at LSB matching, the wavelet filter whose thresholding is improved is used to filter the testing image, and the result is regarded as the estimation of the cover image. Integrating the HCF-COM/AHCF-COM features between the original image and the filtered one, a new method of LSB matching steganalysis is proposed. The experimental results show that, this method can detect not only the JPEG compressed images efficiently, but also uncompressed high-resolution raw scanned images.3. A LSB matching steganalysis method is proposed on the basis of histogram properties. According to the different properties in histogram,difference histogram and subband coefficients histogram after wavelet decomposition between before and after the LSB matching, 23 features are proposed for SVM training, and a classifier based on these features is used to distinguish cover image and stego one.4. A LSB matching steganalysis method is proposed on the basis of deviation degree. At first the linear deviation degree of histogram and the plane deviation degree of co-occurrence matrix are defined, so 10 features are proposed for LSB matching steganalysis. The twice embedded process is used to eliminate the influence of diffirent image contents and improve the performance of steganalysis.Finally, we summarize our research work, and discuss further research topics and directions in the future.