资源描述:
《2008.Adaptive local linear regression with application to printer color management》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、936IEEETRANSACTIONSONIMAGEPROCESSING,VOL.17,NO.6,JUNE2008AdaptiveLocalLinearRegressionWithApplicationtoPrinterColorManagementMayaR.Gupta,Member,IEEE,EricK.Garcia,andErikaChinAbstract—Locallearningmethods,suchaslocallinearregres-regressionestimateisboundedbythevarianceofthem
2、easure-sionandnearestneighborclassifiers,baseestimatesonnearbymentnoise.trainingsamples,neighbors.Usually,thenumberofneighborsWeapplyourproposedadaptivelocallinearregressiontousedinestimationisfixedtobeaglobal“optimal”value,chosenprintercolormanagement.Colormanagementrefersto
3、thetaskbycrossvalidation.Thispaperproposesadaptingthenumberofneighborsusedforestimationtothelocalgeometryofthedata,ofcontrollingcolorreproductionacrossdevices.Manycom-withoutneedforcrossvalidation.Thetermenclosingneighbor-mercialindustriesrequireaccuratecolor,forexample,for
4、thehoodisintroducedtodescribeasetofneighborswhoseconvexproductionofcatalogsandthereproductionofartwork.Inaddi-hullcontainsthetestpointwhenpossible.Itisproventhaten-tion,therisingubiquityofcheapcolorprintersandthegrowingclosingneighborhoodsyieldboundedestimationvarianceunder
5、sourcesofdigitalimageshasrecentlyledtoincreasedconsumersomeassumptions.Threesuchenclosingneighborhooddefinitionsarepresented:naturalneighbors,naturalneighborsinclusive,demandforaccuratecolorreproduction.andenclosingk-NN.TheeffectivenessoftheseneighborhoodGivenaCIELABcolortha
6、tonewouldliketoreproduce,thedefinitionswithlocallinearregressionistestedforestimatingcolormanagementproblemistodeterminewhatRGBcoloronelookuptablesforcolormanagement.Significantimprovementsinmustsendtheprintertominimizetheerrorbetweenthedesirederrormetricsareshown,indicatingt
7、hatenclosingneighborhoodsCIELABcolorandtheCIELABcolorthatisactuallyprinted.maybeapromisingadaptiveneighborhooddefinitionforotherlocallearningtasksaswell,dependingonthedensityoftrainingWhenappliedtoprinters,colormanagementposesaparticularlysamples.challengingproblem.Theoutput
8、ofaprinterisanonlinearfunc-tionthatdependsonavarietyofnontrivialfactors,includingI