资源描述：
《A Framework for Crowdsourced Multimedia Processing and Querying一种众包多媒体处理与查询框架》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、AFrameworkforCrowdsourcedMultimediaProcessingandQueryingAlessandroBozzonIlioCatalloEleonoraCiceriPieroFraternaliDavideMartinenghiMarcoTagliasacchiDipartimentodiElettronicaeInformazioneViaPonzio34/5Milan,Italyfirst.last@polimi.itABSTRACThavebeenexploitedto
2、addressproblemswherehumansout-performmachines,mostnotablycommonsenseknowledgeThispaperintroducesaconceptualandarchitecturalframe-elicitationandcontenttaggingformultimediasearch[3].workforaddressingthedesign,executionandveri cationofInthelatter eld,crowds
3、ourcedmultimediacontentpro-tasksbyacrowdofperformers.Theproposedframeworkcessingexploitsthefactthathumanshavesuperiorcapac-issubstantiatedbyanongoingapplicationtoaproblemofityforunderstandingthecontentofaudiovisualmaterials,trademarklogodetectioninvideoc
4、ollections.Preliminaryandthusreplacestheoutputofautomaticcontentclassi -resultsshowthatthecontributionofcrowdscanimprovethecationwithhumanannotations,andfeatureextractionwithrecallofstate-of-the-arttraditionalalgorithms,withnolosshuman-madetags.intermsof
5、precision.However,task-to-executormatching,Inthispaper,weadoptadi erentapproach:ratherthanasexpected,hasanimportantin
uenceonthetaskperfor-replacingfeatureextractionalgorithms,weaimatimprov-mance.ingtheirperformancewithwell-selectedtasksassignedtohumanex
6、ecutors,thusrealizingamoreintegratedcollabo-CategoriesandSubjectDescriptorsrationbetweenhumanjudgementandalgorithms.H.4[InformationSystemsApplications]:Miscellaneous;Thecontributionofthepaperistheillustrationofthede-D.2.8[SoftwareEngineering]:Metrics
7、com
8、plexitymea-sign,implementation,andpreliminaryevaluationofacrowd-sures,performancemeasuressearchingapplicationfortrademarklogodetectioninvideocollections,inwhichthehelpofthecrowdissoughtforse-GeneralTermslectingthemostappropriateimagesassociatedwithabrand
9、name,soastoimprovetheprecisionandrecallofaclassi-Design,Experimentation,HumanFactors,Measurement,Per-calimageretrievalapproachbasedonlocalfeatures(e.g.formanceSIFT)[5].Thepaperisorganizedasfollows:priortointroducingtheKeyw
