基于主饰品纹理形状特征的人工神经网络蜡染分类(IJISA-V9-N6-6)

I.J.IntelligentSystemsandApplications,2017,6,55-65PublishedOnlineJune2017inMECS()DOI:10.5815/ijisa.2017.06.06Copyright©2017MECSI.J.IntelligentSystemsandApplications,2017,6,55-65BatikClassificationwithArtificialNeuralNetworkBasedonTexture-ShapeFeatureofMainOrnamentAnitaAhmadKasimDepartmentofComputerScienceandElectronics,FacultyofMathematicsandNaturalSciences,UniversitasGadjahMada,Yogyakarta,IndonesiaDepartmentofInformationTechnology,FacultyofEngineering,UniversitasTadulako,IndonesiaE-mail:[e-mail]sHarjokoDepartmentofComputerScienceandElectronics,FacultyofMathematicsandNaturalSciences,UniversitasGadjahMada,Yogyakarta,IndonesiaE-mail:[e-mail],[e-mail]—BatikisatextilewithmotifsofIndonesianculturewhichhasbeenrecognizedbyUNESCOasworldculturalheritage.Batikhasmanymotifswhichareclassifiedinvariousclassesofbatik.Thisstudyaimstocombinethefeaturesoftextureandthefeatureofshapes’ornamentinbatiktoclassifyimagesusingartificialneuralnetworks.Thevalueoftexturefeaturesofimagesinbatikisextractedusingagraylevelco-occurrencematrices(GLCM)whichincludeAngularSecondMoment(ASM)/energy),contrast,correlation,andinversedifferentmoment(IDM).Thevalueofshapefeaturesisextractedusingabinarymorphologicaloperationwhichincludescompactness,eccentricity,rectangularityandsolidity.Atthisphaseofthetrainingandtesting,wecomparethevalueofaclassificationaccuracyofneuralnetworksineachclassinbatikwiththeirtexturefeatures,theirshape,andthecombinationoftextureandshapefeatures.Fromthethreefeaturesusedintheclassificationofbatikimagewithartificialneuralnetworks,itwasobtainedthatshapefeaturehasthelowestaccuracyrateof80.95%andthecombinationoftextureandshapefeaturesproducesagreatervalueofaccuracyby90.48%.Theresultsobtainedinthisstudyindicatethatthereisanincreaseinaccuracyofbatikimageclassificationusingtheartificialneuralnetworkwiththecombinationoftextureandshapefeaturesinbatikimage.IndexTerms—Batik,ArtificialNeuralNetwork,Texture-ShapeFeature.I.INTRODUCTIONBatikisamotifsintextilewithahighartisticvalue,spreadacrossAsia,especiallyinCentralJavaIndonesia[1].Theuniquenessofbatikfabriccomesfromtheproductionprocesscalledmbatik.Theprocessofmbatikproducespatternsthatarediverseandhighineconomicvalues.Batikdesignsareinspiredbynatureormythologyandthereforeresultingtypicalgeometricalpatterns,variousmotifsofbatikwithdifferentnames[2].Thusitmakesitdifficultinclassifyinganypattern.Thepurposeofimageclassificationinbatikistodividebatikimagebasedontheclassofeachpatternthuscaneasilyberecognizedbyitsfeatures.Batikhasastructuremodelconsistingofprimaryandadditionalornaments.Eachmainornamentinbatikhasphilosophicalmeaning,forexample,Grompolornament,worninaweddingceremony.Grompolmeanstogatherortounite,thehopeofcollectingeverythingwasexcellentasfortune,happiness,offspring,livinginharmonyandsoon[3].Additionalornamentationissmallerandsimpleanddoesnothaveaphilosophicalmeaninginthecompositionofbatikpatterns.Inonebatikpatternmaycompriseseveralotherornamentations.Batik’stexturesarediverse.Forinstance,therearetextureswithpatternsoftheedgesofboldlineswhichhaveahighcontrastvalueortheedgesoffuzzylineswhichhavelowcontrastvalues.Regardingthesizeoftheedgesofthelines,therearethickandthin.Meanwhile,thereisalarge,mediumandsmallsizeofthemainbatikornaments[4].ThevariousnumberofbatikpatterncausesdifficultiesinidentifyingthepatternsinIndonesia.Forthat,weneedamethodthatcanclassifyeachbatikpatternbasedonitsmainornamentpattern.Inthispaper,weproposedaclassificationmodelofbatikimagebyusingneuralnetworksbasedontexturefeaturesandshapefeaturesofthemainornaments.Thispaperwilldiscussbatikanditsclassificationsinthefirstpart,whilethesecondpartpresentsresearcheverconductedonbatik.Thethirdsectiondiscussestheresearchmethodsusedandfinally,thefourthandfifthsectionsdiscusstheresearch’sresults,discussion,andconclusions.56BatikClassificationwithArtificialNeuralNetworkBasedonTexture-ShapeFeatureofMainOrnamentCopyright©2017MECSI.J.IntelligentSystemsandApplications,2017,6,55-65II.RELATEDWORKSeveralstudieshavebeenconductedforproperclassificationofbatikimage[1][5][6][7].Oneresearchonbatikwasperformingclassificationbycolor,contrast,andmotifswhichweretheworkbyMoertiniandSitohang[8].TheyusedtheHSVcolormodelforclassifyingbatikbasedonwaveletmethodtoextractthecolorofbatiktexturefeature.Theresultsobtainedshowedaclusteringalgorithmwiththecolorsandtextureslookinggood.Thebatiktexturefeaturescanbeachievedbythemethodofco-occurrencematricesofsub-bandimages.Thismethodcanbeusedtoclassifybatikimagewhichwasrandomlyobtainedfromtheinternet.Thisapproachcombinesthemethodsofgray-levelco-occurrencematrices(GLCM)anddiscretewavelettransform(DWT).Firstofall,theimageiscomposedwithDWTtobecomesub-bandimage.Then,thetexturefeaturesofsub-bandimageareextractedwithGLCM.Thevaluesoftheextractedresultsbecometheinputtotheprobabilisticneuralnetworks(PNN).Theresultsaregoodenoughtoclassifytheimageofbatik.Themaximumaccuracythatcanbeachievedis72%[9].Nugrowati,etal[10]proposedanimageclassificati