资源描述:
《Exploring Cross-Channel Texture Correlation for Color Texture Classification_CCLBP_BMVC2013》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、XIANBIAOQIETAL.:CCLBP1ExploringCross-ChannelTextureCorrelationforColorTextureClassificationXianbiaoQi11BeijingUniversityofPostsandqixianbiao@gmail.comTelecommunications,P.R.ChinaYuQiao22ShenzhenKeyLabofCVPR,yu.qiao@siat.ac.cnShenzhenInstituteofAdvancedChun-GuangLi1Techno
2、logy,P.R.Chinalichunguang@bupt.edu.cnJunGuo1guojun@bupt.edu.cnAbstractThispaperproposesanovelapproachtoencodecross-channeltexturecorrelationforcolortextureclassificationtask.Firstly,wequantitativelystudythecorrelationbetweendifferentcolorchannelsusingLocalBinaryPattern(L
3、BP)asthetexturedescriptorandusingShannon’sinformationtheorytomeasurethecorrelation.Wefindthat(R,G)chan-nelpairexhibitsstrongercorrelationthan(R,B)and(G,B)channelpairs.Secondly,weproposeanoveldescriptortoencodethecross-channeltexturecorrelation.Theproposeddescriptorcancap
4、turewelltherelativevarianceoftexturepatternsbetweendifferentchannels.Meanwhile,ourdescriptoriscomputationallyefficientandrobusttoimagero-tation.Weconductextensiveexperimentsonfourchallengingcolortexturedatabasestovalidatetheeffectivenessoftheproposedapproach.Theexperimen
5、talresultsshowthattheproposedapproachsignificantlyoutperformsitsmostlyrelevantcounterpart(Multi-channelcolorLBP),andachievesthestate-of-the-artperformance.1IntroductionColorandTexturearetwoimportantaspectsofnaturalimages.Itiswidelyrecognizedthattheyprovidestrongcomplemen
6、tarycuestoeachotherinalotofcomputervisionapplica-tions,suchasobjectrecognition[21,26],flowerrecognition[14,19],textureclassification[18],materialrecognition[10,23],content-basedimageretrieval,colortexturesegmentationandmanymore.Recently,LocalBinaryPattern(LBP)descriptor[1
7、5]anditsvariants[6,7,8,16,19,30]havebeenwidelyappliedontexturerelevanttasksduetotheircomputationalefficiencyandgreattexturediscriminativeability.In[16],Timoetal.systematicallyintroducedtheLBPoperatoranditsseveralvariantsincludinguniformLBPandrotationinvariantLBP.Sincethe
8、n,LBPhasbeenwidelyappliedtoalotoftasks,suchasfacerecognition,imageretrieval,andmanymore[17].Recently,in[7],Guo