资源描述：
《sift算法详细讲解(英文经典)》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、DistinctiveImageFeaturesfromScale-InvariantKeypointsDavidG.LoweComputerScienceDepartmentUniversityofBritishColumbiaVancouver,B.C.,Canadalowe@cs.ubc.caJanuary5,2004AbstractThispaperpresentsamethodforextractingdistinctiveinvariantfeaturesfromimagesthatca
2、nbeusedtoperformreliablematchingbetweendifferentviewsofanobjectorscene.Thefeaturesareinvarianttoimagescaleandrotation,andareshowntoproviderobustmatchingacrossaasubstantialrangeofaffinedis-tortion,changein3Dviewpoint,additionofnoise,andchangeinilluminati
3、on.Thefeaturesarehighlydistinctive,inthesensethatasinglefeaturecanbecor-rectlymatchedwithhighprobabilityagainstalargedatabaseoffeaturesfrommanyimages.Thispaperalsodescribesanapproachtousingthesefeaturesforobjectrecognition.Therecognitionproceedsbymatch
4、ingindividualfea-turestoadatabaseoffeaturesfromknownobjectsusingafastnearest-neighboralgorithm,followedbyaHoughtransformtoidentifyclustersbelongingtoasin-gleobject,andfinallyperformingverificationthroughleast-squaressolutionforconsistentposeparameters.Th
5、isapproachtorecognitioncanrobustlyidentifyobjectsamongclutterandocclusionwhileachievingnearreal-timeperformance.AcceptedforpublicationintheInternationalJournalofComputerVision,2004.11IntroductionImagematchingisafundamentalaspectofmanyproblemsincomputer
6、vision,includingobjectorscenerecognition,solvingfor3Dstructurefrommultipleimages,stereocorrespon-dence,andmotiontracking.Thispaperdescribesimagefeaturesthathavemanypropertiesthatmakethemsuitableformatchingdifferingimagesofanobjectorscene.Thefeaturesare
7、invarianttoimagescalingandrotation,andpartiallyinvarianttochangeinilluminationand3Dcameraviewpoint.Theyarewelllocalizedinboththespatialandfrequencydomains,re-ducingtheprobabilityofdisruptionbyocclusion,clutter,ornoise.Largenumbersoffeaturescanbeextract
8、edfromtypicalimageswithefficientalgorithms.Inaddition,thefeaturesarehighlydistinctive,whichallowsasinglefeaturetobecorrectlymatchedwithhighprobabilityagainstalargedatabaseoffeatures,providingabasisforobjectandscenerecognition.Thecostofex
