In　this　paper,　we　propose　an　improved　SIFT　algorithm　based　on　FAST　corner　detection.　We　first　estimate　the　level　of　the　pyramid　by　the　size　of　the　image.　Then,　at　each　level,　we　estimate　the　threshold　by　the　image　content.　By　the　threshold,　we　implement　the　FAST　corner　detection.　At　each　level,　the　FAST　corner　detection,　generation　of　SIFT　descriptors　and　coordinate　mapping　are　implemented　by　Wagner＇s　idea.　The　experimental　results　show　that　our　proposed　algorithm　has　the　similar　matching　performance　and　runs　faster.