搜索
I.J.Image,GraphicsandSignalProcessing,2013,10,1-8PublishedOnlineAugust2013inMECS()DOI:10.5815/ijigsp.2013.10.01Copyright©2013MECSI.J.Image,GraphicsandSignalProcessing,2013,10,1-8ImageIdentificationBasedonShapeandColorDescriptorsandItsApplicationtoOrnamentalLeafKoheiArai,IndraNugrahaAbdullah,HiroshiOkumuraGraduateSchoolofScienceandEngineering,SagaUniversity,Saga,JapanAbstract—Humanhasadutytopreservethenature,preservingtheplantisoneoftheexamples.Thisresearchhasanemphasisonornamentalplantthathasfunctionalitynotonlyasornamentbutalsoasmedicine.AlthoughinIndonesia,ingeneralthisplantiscultivatedinfrontofthehouse;onlyfewpeopleknowaboutitsmedicinalfunction.Consideringthiseasinesstoobtainanditsmedicinalfunction,thisplanthastobeaninitialtreatmentoroptiontowardsfullchemical-basedmedicines.Thisresearchproposesasystemwhichabletoidentifyproperlyornamentalplantfromitsleafutilizingitsshapeorcolorfeatures.ShapedescriptorrepresentedbyDyadicWaveletTransformationandZernikeComplexMoment,andHSV-basedcolorhistogramascolordescriptor.Thisresearchprovidesbenefitofthesethreemethodstosolvevarioustestaspects.Itwasobtained81.77%ofoverallaverage-testingperformance.IndexTerms—Identification,Dyadicwavelet,Zernikemoments,HSV,SVM,leaf,overlappingI.INTRODUCTIONOxygenisanessentialpartforalllivingthingsintheworld.Plantplaysanimportantroletoproduceoxygenandsupplyitfortheirsustainablelife.Thecyclebetweenhumanandtheplantistheinterestingone.Carbondioxideastheoutputofhumanrespiratoryisneededbyplantforphotosynthesisactivity.Then,thisactivityisresultingoxygenwhichvitalforhuman.Accordingtothiscycle,humansupposedtopreservetheplanttomaintainavailabilityofoxygen.BasedonInternationalUnionforConservationofNatureandNaturalResources,thenumberofidentifiedplantspeciesintheworldwhichconsistofMosses(M),FernsandAllies(FA),Gymnosperms,FloweringPlants(FP),RedandGreenAlgae(RGA)isabout307.674species[1].Figure.1ColumnchartsnumberofidentifiedplantspeciesOntheotherside,theapproximatenumberofunidentifiedspeciesis86.429species.ItconsistsofFloweringPlantswith83.400species,FernsandAllieswith3.000species,Mosseswith29species[2].ConsideringthehighestnumberpossessedbyFloweringPlants,identificationoftheplants,whichincludealsoornamentalplant,hasbecomeachallengeforus.Asarecognitionstepoftheunidentifiedspecies,thisresearchhasanemphasisonornamentalleafwhichhasmedicinalfunction.AlthoughmainlyinIndonesiathisplantiscultivatedinfrontofthehouse;onlyfewpeopleknowaboutitsmedicinalfunction.Ifthiseasinessanditsmedicinalfunctionaretakenintoconsideration,thisplanthastobeaninitialtreatmentoroptiontowardsfullchemical-basedmedicines.Identificationofleafimageispossiblydonethroughidentificationofsomeleaffeatures,i.e.shape,venation,andcolor.Nevertheless,mostoftheresearcherswereusingshapefeaturetoidentifytheleaf.Shapeisasubstantialparttodescribeimagecontent.Naturally,welostonedimensionofinformationwhileprojectingofarealworldthree-dimensionalobjectontothetwo-dimensionalprojectionplane.Thisreasonmakestheshapefeaturesareoftenonlyrepresentprojectedobjectnotrealworldobject.Propermethodsasthesolutionofrealworldobjectproblemsareneeded.Therepresentationofshapecanbeseparatedintocontour-basedandregion-based.Bothrepresentationscanbefurtherseparatedintostructuralandglobal[3].DyadicwavelettransformationandZernikecomplexmomentsarerepresentingglobalmethodofcontour-basedapproach,andglobalmethodofregion-based050000100000150000200000250000300000FPMFARGANumberofspecies2ImageIdentificationBasedonShapeandColorDescriptorsandItsApplicationtoOrnamentalLeafCopyright©2013MECSI.J.Image,GraphicsandSignalProcessing,2013,10,1-8approach,respectively.Weareexpectingnicerepresentationofshapeusingacombinationfromcontour-basedandregion-based.Formerly,manyresearchershadconductedimageprocessingusingDyadicwavelettransformation.Asin[4]and[5],theywereapplyingthismethodforimagedenoisingpurpose.Theresultsforbothresearchersweresatisfying.In[7],theywereadoptingZernikecomplexmomentstoclassifybenignandmalignantmassesonMammogramimages.Themomentsweredividedintotwogroups.Firstgroupwaslow-orderofthemoments,andsecondwashigh-orderofthemoments.Itwasconcludedthatlow-ordermagnitudeofZernikecomplexmomentwaspreferredincomparisontohigh-order.Inleafidentificationarea,Park,Hwang,andNam[11]haddevelopedimageretrievalsystemforleafimagescollection.Utilizationofleafvenationbecametheirprimarypointtoidentifytheleaf.Then,followedbytheworkthatproposedbyWanget.al.[13],theyclassifiedleafimagewithcomplicatedbackground.SegmentationprocessoftheleafutilizingautomaticmarkerforWatershedmethod,andClassificationprocesswasdonebycombining7Humomentsand16Zernikemoments.Thisworkinspiresustobefurtherdevelopedinsegmentationandmomentarea.Thecolorfeatureisalsoimportanttoidentifytheleaf.Inreallife,frequently,theidentificationprocessisaffectedbyaspect,i.e.lightingchangedfromthesun.WehopeutilizationofHSVcanguaranteeustoidentifytheleafproperly.Inthispaper,wedescribetheornamentalleafinthesection2,followedbytheproposedmethodforimageidentificationbasedonshapeandcolordesc