JPEG Images Blind Steganography Detection Methods Based On The Feature Fusion Technology

Information steganographic technique is a technology to hide information in a multimedia carrier, but lawbreakers can use the steganography technology to conceal their criminal conspiracy by secret communication, so steganography detection technology is an effective antidote for such wrongful action. The feature fusion technology can effectively discover the complementarity among different feature sets, remove redundant information, and more comprehensively integrate feature classification information in order to effectively improve the classification detection performance of steganography detection. The extensive use of the JPEG images has made it to be the main carrier of the image steganographic techniques, so improving the JPEG images blind steganography detection performance by the feature fusion technology has become a hot research topic in steganography detection field. However, there are some shortcomings in current feature fusion technology:1. There is a certain blindness in selecting what kind of basic feature sets to be fused; 2. Serial feature fusion will lead to a new feature dimension of excessive growth, and may cause the curse of dimensionality; 3. If there are too much difference in the dimensions of the basic feature sets, parallel feature fusion need artificially operation to add many zeros, so too much redundant information will be artificially introduced; 4. The typical parallel feature fusion method by the structure of complex vector will hard to break the limit on the fusion number of features. Therefore, this paper uses the above four shortcomings of feature fusion as the research mainline, and focuses on the feature selection metrics of serial feature fusion, reducing the adding zeros operation in parallel feature fusion and researching three or more multi-feature fusion method, so the specific methods are designed as follows:1. Bring the kernel canonical correlation analysis method to the serial feature fusion technology, and use the kernel canonical correlation coefficient as the standard to measure the redundancy and integration of two JPEG images steganography feature sets, and improve the detection performance.2. Introduce the principal component analysis method to the parallel feature fusion technology, and extracte the principal components of two JPEG images steganography features with different dimensions by principal component analysis, and reduce the gap of dimensions in order to overcome the restriction of excessive growth dimension in serial feature fusion and artificially operation to add many zero in parallel feature fusion, and improve the detection performance and time performance.3. Use the quaternion theory in the parallel features fusion technology, and use principal component analysis to extracte the principal components of four JPEG images steganography features with different dimensions and unify their dimensions, after that use quaternion to parallel fuse four feature sets in order to overcome the limit on the fusion number of the typical complex vector parallel feature fusion, and improve the detection performance.Finally, the paper gives some comparative analysis of this three blind steganography detection feature fusion methods, and integrates them as a JPEG images blind steganalysis system based on the feature fusion technology.