基于多进制小波变换及多维纹理特征的遥感影像融合方法(英文).docx

基于多进制小波变换及多维纹理特征的遥感影像融合方法(英文)
Introduction：
Remotesensingtechnologyhasbecomeanimportantmeansofacquiringspatialinformationontheearth'ssurface,andremotesensingimagefusionhasbecomeanessentialpartofremotesensingimageprocessing.Remotesensingimagefusionaimstomergemultipleimagesofthesamesceneintoafusedimagetomakefulluseofthecomplementarybenefitsofmultipleimages.However,thetraditionalmethodsofremotesensingimagefusion,suchastheintensity-basedmethods,havesomeshortcomingsinpreservingthespectralandspatialinformationoftheoriginalimages.Thispaperproposesaremotesensingimagefusionmethodbasedonmulti-radixwavelettransformandmultidimensionaltexturefeaturestoaddresstheseissues.
Theory：
Themulti-radixwavelettransformcanextractfeaturesofdifferentscalesanddirectionsinremotesensingimages.Thepaperusesthistransformtodecomposethesourceimagesintolow-frequencyandhigh-frequencysub-bands.Then,thelow-frequencysub-bandsofthesourceimagesarefusedbythemaximumselectionrule.Thehigh-frequencysub-bandsarefusedbytheNSST(NonsubsampledShearletTransform)algorithm,whichcaneffectivelypreservethedetailinformationoftheoriginalimagesandenhancethespectralinformation.Finally,thefusedimageisreconstructedusingtheinversewavelettransform.
Inadditiontothemulti-radixwavelettransform,thispaperalsoproposesamethodoffeaturecomplementationbasedonmultidimensionaltexturefeatures.ThemethodusestheGaborfiltertoextracttexturefeaturesofremotesensingimagesindifferentdirections,scales,andfrequencies.Then,thetexturefeaturesofthefusedimagearecombinedwiththetexturefeaturesoftheoriginalimagesbythemaximumselectionrule.
Experiments：
Theproposedmethodwasevaluatedonthreeremotesensingimagedatasets,includingQuickBird,IKONOS,andLandsat7dataset.Thequalitativeandquantitativeresultsshowthattheproposedmethodoutperformsthetraditionalmethodsintermsofspectralandspatialinformationpreservation,imagecontrastenhancement,andvisualquality.
Conclusion：
Thispaperproposesaremotesensingimagefusionmethodbasedonmulti-radixwavelettransformandmultidimensionaltexturef