The　width　of　urban　road　is　various.　Roads　with　different　width　have　different　appearance　in　an　image.　Multi-resolution　analysis-based　method　for　road　extraction　was　proposed.　Firstly,　some　wide　roads　in　IKONOS　multi-spectral　imagery　were　extracted　through　using　level　setbased　method.　While　some　narrow　roads,　which　are　non-salient　in　4m　multi-spectral　imagery,　were　extracted　from　1m　fusion　imagery.　Because　the　central　green　belt　and　some　continuously　arranged　road-side　trees　can　imply　a　road　existence,　this　vegetation　information　has　also　been　extracted,　which　effectively　assist　the　road　extraction.　Finally,　the　accuracy　of　the　road　extraction　result　was　quantitatively　assessed.　The　assessment　result　shows　that　this　proposed　method　has　a　good　performance.　This　method　not　only　can　effectively　extract　salient　road　network　from　urban　image,　but　also　can　extract　some　non-salient　narrow　roads.　©　2016　IEEE.