This　paper　presents　a　multistage　framework　for　road　extraction　from　remotely　sensed　imagery.　The　framework　is　a　combination　of　multifeature-based　mean　shift,　SVM　classifier,　and　shape　feature　filter.　Mean　shift　is　first　employed　to　cluster　the　remotely　sensed　images　in　the　fusion　of　spectral-texture　features　and　then　a　classifier　of　SVM　is　applied　to　classify　the　segmented　images　into　two　classes:　the　road　class　and　the　non-road　class.　Finally,　the　road　class　is　refined　by　using　road　shape　features.　The　experimental　results　show　that　this　method　is　efficient　in　extracting　road　from　remotely　sensed　imary.　©　2016.