As　an　important　foundation　for　image-guided　technology,　image　matching　technique　is　the　key　technology　of　modern　war.　This　paper　proposes　a　new　algorithm　of　affine　invariant　detector　and　descriptor　of　local　invariant　feature　points,　starting　from　feature　point　detection　and　description　point　of　view,　making　up　the　traditional　feature　point　extraction　defects　of　small　number　and　types.　Meantime,　proposes　an　improved　similarity　measure　method　based　on　the　previously　proposed　new　feature　point　detection　and　description　algorithm,　it　improves　the　matching　accuracy　and　real-time　performance.　Finally,　compares　the　experiment　results　of　SURF,　SIFT　and　the　improved　algorithm　proposed　in　this　paper,　the　experimental　results　shows　that　the　feature　points　extracted　by　the　improved　algorithm　has　fully　affine　invariance,　and　improved　the　accuracy　and　speed　of　image　matching　algorithm　efficiently.