In　order　to　track　facial　feature　points,　according　to　the　different　features　of　different　points,　many　methods　are　used　to　track.　For　the　points　on　the　brow,　the　authors　use　optical　flow　to　track,　and　the　authors　also　use　Laplacian　pyramid　image　replacing　gray　image　as　the　input　image　to　improve　the　tracking　veracity.　For　the　points　on　the　mouth,　the　authors　combine　the　optical　flow　with　EBGM,　using　optical　flow　to　get　the　position　roughly　and　reduce　searching　area　to　accelerate　the　tracking　speed.　For　the　points　on　the　eyelids,　image　binarization　method　is　used.　Then　based　on　MPEG-4　mechanism,　the　authors　clone　the　tracked　motion　data　to　an　exaggerated　character,　which　is　automatically　generated　by　the　system　according　to　a　real　face.　Experiments　show　that　the　result　is　recreational.