The　model　of　visual　attention　proposed　by　Stentiford　can　effectively　identify　salient　regions　of　an　image.　Since　Stentiford　model　is　produced　by　using　evolutionary　programming　(EP),　which　usually　involves　a　lot　of　computation,　a　fast　region　of　interest　(ROI)　extraction　approach　is　proposed.　The　input　image　is　firstly　bilinear　down-sampled.　The　visual　attention　map　is　calculated　with　occurrence　frequency　of　a　region　in　the　image.　The　candidate　ROI　is　obtained　after　filling　holes　and　filtering　isolated　points.　The　actual　ROI　is　extracted　by　scaling　the　candidate　ROI　with　bilinear　interpolation.　Experimental　results　show　that　the　proposed　approach　can　faster　extract　ROI,　and　have　a　more　accurate　ROI　detection　as　well.　©　2010　IEEE.