不同来源织物图像的人工神经网络分类方法(IJIGSP-V11-N12-4)

I.J.Image,GraphicsandSignalProcessing,2019,12,29-38PublishedOnlineDecember2019inMECS()DOI:10.5815/ijigsp.2019.12.04Copyright©2019MECSI.J.Image,GraphicsandSignalProcessing,2019,12,29-38ANNApproachforClassificationofDifferentOriginFabricImagesBasavarajS.AnamiK.L.E.InstituteofTechnology,Hubballi,580030,IndiaEmail:[e-mail]eeringandResearch,Belagavi,590008,IndiaEmail:[e-mail]:26July2019;Accepted:23August2019;Published:08December2019Abstract—Thispaperfocusesonclassificationofvarietiesofplants’,animals’andminerals’originfabricmaterialsfromimages.Themorphologicaloperations,namely,erosionanddilationareused.ANNclassifierisusedtopredicttheclassificationratesandtheratesof89%,87%and88%areobtainedforplants’,animals’andminerals’originfabricimagesrespectively.Theoverallclassificationrateof88%isobtained.IndexTerms—Morphology,Plantorigin,Animalorigin,Mineralorigin,Featureextraction,ANN.I.INTRODUCTIONImageprocessingisaninterdisciplinaryfieldusedintheautomaticextraction,analysisandunderstandingofusefulinformationfromimages.Itisusedindifferentfields,namely,agriculture,industries,biologicalfield,medicalfieldandthelike.ThetextileindustryplaysmostimportantroletoIndianeconomythroughitsexportearnings.Indiaistheworld’ssecondlargestproduceroftextilesintheworld.Thedifferentformsoftextilesareagro-textiles,geo-textiles,fabrics,medical-textilesetc.Indiahasgoodresourcesoffibers,suchascotton,wool,polyester,silk,fur,jute,denim,etc.Thedifferentfabricmaterialshavedifferenttextures,colorsandmorphologicalfeatures.Thefabricmaterialsavailableatpresentareofdifferentorigins,suchasplant,animalandmineral.Plantoriginfabrichasanaturaldimcolorwithcoarserfinish,wherein,theanimaloriginfabricisbrightincolorandhasclosedview.But,themineraloriginfabricmaterialhasfinerfinishandclosedview.Themineralfabricimageshaverefractiveinnatureandlookbrighterincolor.Hence,thefabricimageswithdifferentoriginshavetheirownuniquevisualfeaturesandthesameareexploredtoclassifythemintotheirrespectivetypesandvarieties.Intheworkcarriedout,fourvarietieseachofplants’,animals’andminerals’originfabricimagesareconsideredandsampleimagesofvarietiesofdifferentoriginfabricareshowninFig.1,Fig.2andFig.3respectively.(a)30ANNApproachforClassificationofDifferentOriginFabricImagesCopyright©2019MECSI.J.Image,GraphicsandSignalProcessing,2019,12,29-38(b)(c)(d)Fig.1.Varietiesofplants’originfabric(a)Cotton,(b)Denim(c)Juteand(d)Flax(a)(b)ANNApproachforClassificationofDifferentOriginFabricImages31Copyright©2019MECSI.J.Image,GraphicsandSignalProcessing,2019,12,29-38(c)(d)Fig.2.VarietiesofAnimals’originfabric(a)Fur(b)Leather,(b)Silkand(c)Wool(a)(b)(c)32ANNApproachforClassificationofDifferentOriginFabricImagesCopyright©2019MECSI.J.Image,GraphicsandSignalProcessing,2019,12,29-38(d)Fig.3.VarietiesofMinerals’originfabric(a)Fibre,(b)Glass,(c)Nylonand(d)Polyester.TheclassificationiscarriedoutintwolevelsasshownFig.4.Inthefirstlevel,thefabricimagesareclassifiedintothreetypes,namely,plants’origin,animals’originandminerals’origin.Inturn,theplants’originfabricimagesareclassifiedintofourvarieties,namely,cotton,denim,juteandflax.Further,theanimals’originfabricimagesareclassifiedintofourvarietiesasfur,leather,silkandwool.Finally,theminerals’originfabricimagesarecategorizedintofourvarietiesasfibre,glass,nylonandpolyester.Fig.4.Varietiesofdifferentoriginfabricimages.TheremainingpartofthepaperisorganizedintoFOURsections.SectionIIcontainstheliteraturesurveycarriedoutrelatedtotheproposedwork.SectionIIIgivesproposedmethodologyconsistingofpre-processing,morphologicalfeatureextraction,featureselectionusingFFSTandANNclassifiertoclassifythefabricimages.SectionIVpresentstheresultsanddiscussion.SectionVcontainstheconclusionoftheworkcarriedout.II.LITERATURESURVEYToknowstateoftheartinthetextileindustry,wehaveconductedaliteraturesurveyandfollowingisthegistofpaperscited.WangXinetal.[1]haveproposedaworkonfabricidentificationusingconvolutionalneuralnetwork,Animage-basedfabricretrievaltechniqueisdevelopedtohelpnewfabricstomanageproducts.Thenetworkistrainedwithhugedatasetcontainingdifferentyarn-dyedfabricpatterns.Itisshownthat,theperformanceismaintainedwell,whensimplerdeeparchitectureisused.Butthesameisfoundtodecreasequickly,ifthecontentsoffabricimagesareminimized.ZhongXiangetal.[2]havepresentedapaperonvisionbasedportableyarndensitymeasuremethodforsinglecolorwovenfabricmaterials.Thefabricimagesarecollectedmanuallyusingasmart-phone.AdiscreteFouriertransformisusedtocomputethedensityofthewovenfabric.Theresultsobtained,claimthatthesystemisrobusttomeettherequirementsinfabricindustry.AnamikaSinghetal.[3]haveproposedamethodonfacedetectionandeyesextractionusingSobeledgedetectionusingdifferentmorphologicaloperations.Themethodologyusesthreephases,namely,pre-processing,faceidentification,andeyesextraction.Theworkcarriedoutisfoundtogivegoodclassificationrate.AmeliaCarolinaSparavigna[4]haspresentedaworkonimagesegmentationappliedtotheanalysisoffabrictext