Steganalysis Feature Design Based On Multi-threshold Local Binary Pattern And Co-occurrence Matrix

With the development of computer technology and the Internet,when advanced technology brings more convenience and pleasure to people's daily life,there are also many problems such as information leakage and theft.This makes people have a higher requirement for information content security.Steganography and steganalysis are two important parts of information hiding technology and the research about them have very important academic significance and application value.Digital image is one of the most widely used multimedia in modern computer technology,so digital image steganography and steganalysis is also a very important branch of information hiding.It is always a key problem in the field of information hiding that it is difficult to design a feature extraction method with a high detection rate.In this thesis,we focus on the feature extraction method in the texture and noise region,which is the significant embedding region for the content adaptive steganographic algorithm.The main contributions of this thesis are as follows:1)A set of derivative filters which is adapted for digital image steganalysis tasks is designed.In the design of filters,we derive different order mixed partial derivative matrix according to the mathematical derivative relationship,and select ten as the original derivative filters in the appropriate size.Then according to the direction and symmetry,we make padding and rotated operations on part of the original derivative filters.Next we select some rotation and non-rotation filters form a plurality of linear and nonlinear filters and so as to form the final derivative filter group.Such filters used in the calculation of residuals.2)A feature extraction method based on local binary pattern and co-occurrence matrix is proposed.Local binary pattern is well used in the detection of image texture and co-occurrence matrix is widely used in the steganalysis.In the feature extraction method first calculate residuals,then calculate the multi scale local binary patterns,and next extract the second order co-occurrence matrix,finally the feature is obtained by an appropriate aggregation and non-linear mapping.The experimental results show that such steganalytic feature based on local binary pattern and co-occurrence matrix has better performance than main stream steganalytic feature SRM(Spatial Rich Model).3)For the detection of subtle changes from steganography embedding,a multi threshold local binary pattern(mt LBP)is proposed.Since the definition of original local binary pattern does not have the ability to distinguish a large part of steganography changes,we design a local binary pattern based on the difference of the pixels' value,and apply it to the feature extraction method.The experiments show that the proposed method has a great improvement compared with SRM for many kinds of content adaptive steganography algorithm with different payloads.