Abstract: this　paper 　presents 　an　unsupervised 　se gmentation　method　based　on　the 　feature 　decomposition 　used 　for 　extracting　residential 　areas 　in　hi gh　resolution 　remote 　sensing　image．the　method　was　realized 　by　multi－scale 　feature 　anal ysis　with　wavelet　decompositionconstituting　the 　feature 　space 　from differences 　of 　the 　internalexternal 　structure 　in 　residential 　areas 　and　the 　average　spectral 　radiant 　intensit ymaking　a　self－adaptive 　se gmentation　by　the 　constraint　mean　shift 　al gorithm　to 　texture 　featuresand　achieve 　the 　automatic　extraction　of 　residential 　areas．experiment 　results 　show　that 　the　proposed　method　can　eliminate 　the 　influence 　from　over－detailed 　images 　and　other 　factors 　caused　by　hi ghresolution　on　the 　extraction　of 　residential 　areasand　thus 　extract 　residential 　areas　more　effectivel y．