Automatic　image　annotation　is　a　significant　and　challenging　problem　in　pattern　recognition　and　computer　vision.　Existing　models　did　not　describe　the　visual　representations　of　corresponding　keywords,　which　would　lead　to　appearing　plenty　of　irrelevant　annotations　in　final　annotation　results.　These　annotations　did　not　relate　to　any　part　of　images　considering　visual　contents.　We　propose　a　new　automatic　image　annotation　model　(NAVK)　based　on　relevant　visual　keywords　to　overcome　above　problems.　Our　model　focuses　on　non-abstract　words.　First,　we　establish　visual　keyword　seeds　of　each　non-abstract　word,　and　then　a　new　method　is　proposed　to　extract　visual　keyword　collections　by　using　corresponding　seeds.　Second,　we　propose　adaptive　parameter　method　and　fast　solution　algorithm　to　determine　similarity　thresholds　of　each　keyword.　Finally,　the　combinations　of　above　methods　are　used　to　improve　annotation　performance.　Experimental　results　verify　the　effectiveness　of　proposed　image　annotation　model.　©　2015　Taylor　&　Francis　Group,　London.