To　realize　blind　steganalysis,　image　quality　metrics　(IQMs)　were　introduced　to　measure　statistical　differences　between　original　image　and　its　distortion　version.　These　IQMs　were　sensitive　to　hidden　message　to　be　extracted　as　features　of　images.　Kernel-based　Support　Vector　Machine　(Kernel-SVM)　was　chosen　as　classifier.　Experiment　results　showed　that　this　method　could　reach　a　high　testing　rate　of　hidden　message　of　images.