资源描述:
《dynamic clustering using particle swarm optimization with application in unsupervised imagenew》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、PROCEEDINGSOFWORLDACADEMYOFSCIENCE,ENGINEERINGANDTECHNOLOGYVOLUME9NOVEMBER2005ISSN1307-6884DynamicClusteringusingParticleSwarmOptimizationwithApplicationinUnsupervisedImageClassificationMahamedG.H.Omran,AndriesPEngelbrecht,andAyedSalmanHowever,hierarchic
2、alclusteringtechniquessufferfromtheAbstract—Anewdynamicclusteringapproach(DCPSO),basedfollowingdrawbacks:onParticleSwarmOptimization,isproposed.ThisapproachisñTheyarestatic,i.e.datapointsassignedtoaclusterappliedtounsupervisedimageclassification.Thepropo
3、sedapproachcannotmovetoanothercluster.automaticallydeterminesthe"optimum"numberofclustersandñTheymayfailtoseparateoverlappingclustersduetoasimultaneouslyclustersthedatasetwithminimaluserinterference.lackofinformationabouttheglobalshapeorsizeoftheThealgor
4、ithmstartsbypartitioningthedatasetintoarelativelylargenumberofclusterstoreducetheeffectsofinitialconditions.Usingclusters.binaryparticleswarmoptimizationthe"best"numberofclustersisselected.ThecentersofthechosenclustersisthenrefinedviatheK-Ontheotherhand,
5、partitionalclusteringalgorithmsmeansclusteringalgorithm.Theexperimentsconductedshowthatpartitionthedatasetintoaspecifiednumberofclusters.Thesetheproposedapproachgenerallyfoundthe"optimum"numberofalgorithmstrytominimizecertaincriteria(e.g.asquareerrorclus
6、tersonthetestedimages.function)andcanthereforebetreatedasoptimizationproblems.TheadvantagesofhierarchicalalgorithmsaretheKeywords—ClusteringValidation,ParticleSwarmOptimization,UnsupervisedClustering,UnsupervisedImageClassification.disadvantagesofthepart
7、itionalalgorithmsandviceversa.PartitionalclusteringtechniquesaremorepopularthanI.INTRODUCTIONhierarchicaltechniquesinpatternrecognition[2],hence,thispaperwillconcentrateonpartitionaltechniques.ATAclusteringistheprocessofidentifyingnaturalDgroupingsorclus
8、ters,withinmultidimensionaldata,Partitionalclusteringaimstooptimizeclustercenters,aswellasthenumberofclusters[10].Mostclusteringbasedonsomesimilaritymeasure(e.g.Euclideandistance)algorithmsrequirethenumberofclusterstobespe