Image　mosaicing　technique　is　used　in　creating　vast　scene　images　and　panoramic　images.　It　has　wide　application　value　in　the　cultural　heritage　protection　and　virtual　reality　field.　An　automatic　algorithm　based　on　feature　points　of　images　is　presented　in　this　paper.　Firstly,　feature　points　are　extracted　using　Harris　corner　detector　to　perform　correlation.　Then,　a　new　method　of　filtering　for　matching　points　is　presented　by　choosing　pairs　of　correlation　feature　points　with　clustering　algorithm　before　purifying　them　with　RANSAC　algorithm.　After　projection　transition　matrix　is　worked　out,　images　are　projected　to　a　cylinder　surface.　Finally,　in　order　to　deal　with　ghosting,　a　fast　method　including　optimization　path　and　luminance　weighting　function　is　used　to　obtain　a　good　result.　Experimental　results　show　that　this　is　a　robust　image　mosaics　algorithm,　and　after　choosing　data　with　clustering　algorithm　beforehand,　the　efficiency　is　greatly　enhanced　and　makes　the　algorithm　more　suitable　in　practice.　©　2007　IEEE.