Digital image is one of the important carrier for information hiding. SinceJPEG images have been widely used in our daily life,they are widely used assteganographic carriers, so the research on steganalysis technology for JPEGimages has important application and theoretical value. In the process of imageuniversal blind detection, classifiers are usually used to classify images, so thechoice of classifiers is also one of the important issues in the image universal blinddetection,the advantages and disadvantages of classifier are directly related to thedetection result. The traditional classifiers often have many defects,so thisthesis,from the point of the choice of classifiers, applies several new classifiers inthe field of pattern recognition for universal blind detection and these newclassifiers make good performance in universal blind detection.The major works of this thesis are as follows:Firstly, we have experimentally studied on SVM-based universal steganalysismethod for JPEG images,which is proposed by Fridrich.Secondly, we use random forests to replace traditional classifiers, which haveproblems:over-fitting,low accuracy,inefficiency and so on,for JPEG imagessteganalysis. Experimental results show that:this method can make goodperformance for images steganalysis.Lastly, In this thesis,we use sparse representation classifier for JPEG imagessteganalysis and present a JPEG images universal blind detection method based onsparse representation. Experimental results show that:this method can make goodperformance for images steganalysis. |