Feature　fusion　can　effectively　improve　the　steganographic　detection　capability,　but　the　previous　researches　of　feature　fusion　in　JPEG　image　steganography　detection　rarely　considered　the　nonlinear　correlation　of　features.　This　paper　analyzes　the　correlation　of　JPEG　image　steganographic　features　and　fuses　features　with　lowest　correlation　to　obtain　better　detection　capability　based　on　KCCA　(Kernel　canonical　correlation　analysis),　which　has　a　good　ability　of　nonlinear　correlation　analysis　and　can　eliminate　the　redundancy　of　information　between　features.　Firstly,　analyze　the　＇DCT　extended　feature＇　and　the　＇markov　reduced　feature＇　which　are　classic　features,　and　the　newly　proposed　＇DCT　adaptive　feature＇　in　2011.　Secondly,　select　two　features　with　lowest　correlation　among　them　for　KCCA　feature　fusion.　Finally,　carry　out　experimental　contrasts　with　other　related　methods.　The　experimental　results　show　that　the　proposed　method　is　reasonable　and　effective.　©　2012　IEEE.