The　Markov　extended　features　extraction　performs　well　in　JPEG　image　steganalysis.　The　dimensionality　of　the　feature　space　is　324.　However,　the　high-dimensional　feature　space　does　some　side-effects　to　classifiers.　In　this　paper,　we　combine　the　forward　selection　algorithm　with　F-score　method　to　select　the　Markov　extended　features.　We　then　compress　those　selected　features　to　get　a　smaller　feature　set　according　to　their　directions.　Therefore,　the　dimensionality　of　feature　space　is　reduced　from　324　to　26.　The　experimental　results　are　presented　to　demonstrate　that　our　proposed　scheme　decreases　complexity　of　classifiers＇　training　but　maintaining　the　correct　classification　rate.　©2009　IEEE.