Generally,　the　distance　metric　learning　method　using　sample　classifier　(such　as　Euclidean　distance　computing)　can　not　achieve　perfect　classification　performance　for　the　image　verification.　Nevertheless,　the　random　forests　distance　method　(RFD)　can　overcome　the　shortcoming　of　distance　metric　learning　since　it　can　handle　the　heterogeneous　data　well.　In　addition,　the　distance　metric　learning　method　can　reduce　the　training　time　of　RFD　because　it　can　remove　the　data　correlation.　Therefore　this　paper　proposes　a　fusion　method　of　distance　metric　learning　and　random　forests　distance.　We　obtain　a　matrix　M　using　the　distance　metric　learning　method　and　use　it　to　linearly　transform　the　sample　space,　then　we　classify　new　samples　by　RFD.　We　experiment　on　LFW,　Pubfig　and　ToyCars　datasets　and　the　results　show　that　our　proposed　fusion　method　outperforms　the　single　distance　metric　learning　method　or　RFD　in　the　recognition　accuracy;　the　training　time　of　RFD　is　much　less　in　the　transformed　sample　space.　©　2014　IEEE.