资源描述：
《特征空间稳健性分析：彩色图像分割 毕业论文外文翻译》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、（1）外文文献原文RobustAnalysisofFeatureSpaces:ColorImageSegmentationAbstractAgeneraltechniquefortherecoveryofsignificantimagefeaturesispresented.Thetechniqueisbasedonthemeanshiftalgorithm,asimplenonparametricprocedureforestimatingdensitygradients.Drawbacksofthecurrentmethods(includingrobustclus
2、tering)areavoided.Featurespaceofanynaturecanbeprocessed,andasanexample,colorimagesegmentationisdiscussed.Thesegmentationiscompletelyautonomous,onlyitsclassischosenbytheuser.Thus,thesameprogramcanproduceahighqualityedgeimage,orprovide,byextractingallthesignificantcolors,apreprocessorforco
3、ntent-basedquerysystems.A512512colorimageisanalyzedinlessthan10secondsonastandardworkstation.Graylevelimagesarehandledascolorimageshavingonlythelightnesscoordinate.Keywords:robustpatternanalysis,low-levelvision,content-basedindexing1IntroductionFeaturespaceanalysisisawidelyusedtoolforsol
4、vinglow-levelimageunderstandingtasks.Givenanimage,featurevectorsareextractedfromlocalneighborhoodsandmappedintothespacespannedbytheircomponents.Significantfeaturesintheimagethencorrespondtohighdensityregionsinthisspace.Featurespaceanalysisistheprocedureofrecoveringthecentersofthehighdens
5、ityregions,i.e.,therepresentationsofthesignificantimagefeatures.Histogrambasedtechniques,Houghtransformareexamplesoftheapproach.Whenthenumberofdistinctfeaturevectorsislarge,thesizeofthefeaturespaceisreducedbygroupingnearbyvectorsintoasinglecell.Adiscretizedfeaturespaceiscalledanaccumulat
6、or.Wheneverthesizeoftheaccumulatorcellisnotadequateforthedata,seriousartifactscanappear.TheproblemwasextensivelystudiedinthecontextoftheHoughtransform,e.g..Thus,forsatisfactoryresultsafeaturespaceshouldhavecontinuouscoordinatesystem.Thecontentofacontinuousfeaturespacecanbemodeledasasampl
7、efromamultivariate,multimodalprobabilitydistribution.Notethatforrealimagesthenumberofmodescanbeverylarge,oftheorderoftens.Thehighestdensityregionscorrespondtoclusterscenteredonthemodesoftheunderlyingprobabilitydistribution.Traditionalclusteringtechniques,canbeusedforfeatu
