快速Mean-shift聚类
Paci cGraphics2010
P.Alliez,K.Bala,andK.Zhou(GuestEditors)
Volume29(2010),Number7
Ef cientMean-shiftClusteringUsingGaussianKD-Tree
ChunxiaXiao
MengLiu
TheSchoolofComputer,WuhanUniversity,Wuhan,430072,China
Abstract
Meanshiftisapopularapproachfordataclustering,however,thehighcomputationalcomplexityofthemeanshiftprocedurelimitsitspracticalapplicationsinhighdimensionalandlargedatasetclustering.Inthispaper,weproposeanef cientmethodthatallowsmeanshiftclusteringperformedonlargedatasetcontainingtensofmillionsofpointsatinteractiverate.Thekeyinourmethodisanewschemeforapproximatingmeanshiftprocedureusingagreatlyreducedfeaturespace.Thisreducedfeaturespaceisadaptiveclusteringoftheoriginaldataset,andisgeneratedbyapplyingadaptiveKD-treeinahigh-dimensionalaf nityspace.Theproposedmethodsigni cantlyreducesthecomputationalcostwhileobtainingalmostthesameclusteringresultsasthestandardmeanshiftprocedure.Wepresentseveralkindsofdataclusteringapplicationstoillustratetheef ciencyoftheproposedmethod,includingimageandvideosegmentation,staticgeometrymodelandtime-varyingsequencessegmentation.
CategoriesandSubjectDescriptors(accordingtoACMCCS):I.4[Computingmethodologies]:ImageProcessingandComputerVision—Applications
1.Introduction
Meanshiftisawellestablishedmethodfordatasetclus-tering,ithasbeenwidelyusedinimageandvideoseg-mentation[CM02][WTXC],objecttracking[CRM03],im-agedenoising[BC04],imageandvideostylization[DS02][WXSC04],andvideoediting[WBC 05],italsohasbeenextendedtogeometrysegmentation[YLL 05]and3Dre-construction[WQ04].Meanshiftworksbyde ningaGaus-siankerneldensityestimateforunderlyingdata,andclusterstogetherthepointsthatconvergetothesamemodeundera xed-pointiterativescheme.Althoughmean-shiftworkswellfordataclusteringandobtainpleasingclusteringre-sults,however,thehighcomputationalcomplexityistheoneofmaindif cultiestoapplymeanshifttoclusterlargedataset,especiallyforthosesituationswheretheinteractiveandevenrealtimeclusteringprocessingarepreferred.
ThecomplexityforthestandardmeanshiftprocedureisO(τdn2),wherenisthenumberofthedatapoints,theτisthenumberoftheiterationsformeanshiftcluster-ingprocedure,anddisthedimensionofthepoint.Themostexpensivecomputingisto ndtheclosestneighbor-hoodforeachpointinthedataspace,whichisamultidi-mensionalrangesearchingmethod.Evenusingoneofthe
c2010TheAuthor(s)
c2010TheEurographicsAssociationandBlackwellPublishingLtd.Journalcompilation
PublishedbyBlackwellPublishing,9600GarsingtonRoad,OxfordOX42DQ,UKand350MainStreet,Malden,MA02148,USA.
mostpopularnearestneighborsearchmethod,theANNmethod[AMN 98],givenaquerypointqandε>0,a(1+ε)approximatenearestneighborofqmustbecom-putedinO(cd,εlogn)time,wherecd,εisafactordepend-ingondimensiondandε.Therefore,whenprocessinglargedatasets,thehightimecomplexityleadstoseriousdif culty.Althoughmanyaccelerationtechniqueshavebeenproposed[EDD03,GSM03,YDGD03,WBC 05,CP06,PD07,FK09],furtherimprovementsarestilldesirableforbothperfor-manceandclusteringquality.
Inthispaper,inspiredbythefasthigh-dimensional lter-ingmethodusingGaussianKD-trees[AGDL09],wepro-poseanef cientparadigmformean-shiftprocedurecom-puting.Ourmethodisbasedonfollowingkeyobservation,sincethemeanshiftprocedureclustersthosepointsthatarefeaturesimilar,whiletherearemanyclustersofpointswhicharehighsimilarinfeature,itiswastefultoperformmeanshiftprocedureforeachpointtoconvergetothemode.Thuswe rstclustertheoriginalpointsetintoclustersbasedonfeaturesimilarityusingKD-tree[AGDL09],andobtaintheGaussianweightedsamplesoftheoriginaldataset,whichisthereducedfeaturespaceforapproximatingoriginalpointset.Theninsteadofcomputingmeanshiftdirectlyonorigi-
Answers Final Exam CHEM 3080 April 9, 2012, 2 pm-5 pm CB 121 Name ... no additional paper, but you are allowed to use the back of the pages...
Academic paper Final Paper Literature review ? including the backgrand, previous research limitation, purpose, method, findings and results, futhur research, ...
final_draft_其它语言学习_外语学习_教育专区。The Final Draft Reflecting on ...8 ? by 11 inch white paper ? double spaced, size 12 font ? one inch...
120703_UNIQLO PR Plan CN final-14时 (NXPowerLite)_销售/营销_经管营销_...paper (release) ● 媒体/政府关系 ● 准备并进行面对媒体的准备 ● 紧急时...
Appendix Final exame paper of systemic anatomy for foreign studnts in WZMC Part One: 5.5 Years MBBS Major, 1 Semester of 2006-2007, Systemic Anatomy...
Tegra_Multiprocessor_Architecture_white_paper_Final_v1.1 tegra2 nvidiategra2 nvidia隐藏>> Whitepaper NVIDIA Tegra Multi-processor Architecture V1.0 P age...
1000-1220(2010)07- An approach for Building Complex Network of Web Service...In this paper, the approach of building complex network of web service is...
English_Final_Examination_Paper_A_英语考试_外语学习_教育专区。山东建筑大学往年期末及答案College English Achievement Test (Band I ) 试卷类型 A Paper One ...
take-home paper final_英语学习_外语学习_教育专区 暂无评价0人阅读0次下载举报文档 take-home paper final_英语学习_外语学习_教育专区。Yintong Liu ...
西门子面试题 A6_Answers_to_paper_testing_Final_IT/计算机_专业资料。A6答案Answer to paper testing of SCM Excellence Graduate Program 2011 I. 翻译 1. ...
热门文档