Abstract: Aimingattheproblemthatthetraditionalimagesegmentationalgorithmscannotbeappliedtohighresolutionremotesensingimageswithmanyfeatures（spectral，textureandgeometricfeatures）,aremotesensingimagesegmentationmethodbasedonmulti-featurewasproposed.Thealgorithmintegratedtheimprovedmeanshiftfilteringandautomaticmarkerwatershedtoachievethesegmentationperformance.Firstly，anautomaticmarkerwatershedmethodwasusedtosegmenttheremotesensingimageforextractinggeometricfeatureusingaffinemomentinvariantsofshapeoperator.Secondly，agraylevelco-occurrencematrixofthefirstprincipalcomponentwascalculatedastexturalfeature.Thirdly，multi-featurefilteringresultswereobtainedbyusingimprovedmeanshiftalgorithmincludingspectralfeature.Finally，thefilteringresultswereperformedusingtheautomaticmarkerwatershedsegmentationmethod.Inordertoshowtheeffectoftheproposedmethod，anunsupervisedevaluationandcomparisonoftheimagesegmentationfromtheproposedalgorithmandsinglewatershedsegmentationwereimplementedusingmulti-spectralinformationentropy.Theexperimentalsegmentationresultsshowthattheproposedalgorithmcanreducetheover-segmentationphenomenonefficientlyanditissuitedforthesegmentationofhigh-resolutionmulti-spectralremotesensingimage.

Key words: meanshift, multi-feature, watershedtransform, high-resolutionremotelysensedimagery, imagesegmentation