摘要
Alotof3Dshapedescriptorsfor3Dshaperetrievalhavebeenpresentedsofar.Thispaperproposesanewmechanism,whichemploysseveralexistingglobalandlocal3Dshapedescriptorsasinput.Withthesparsetheory,somedescriptorswhichplaythemostimportantroleinmeasuringsimilaritybetweenquerymodelandthemodelinthedatasetareselectedautomaticallyandanaffinitymatrixisconstructed.Spectralclusteringmethodcanbeimplementedtothisaffinitymatrix.Spectralembeddingofthisaffinitymatrixcanbeappliedtoretrieval,whichintegratingalmostalltheadvantagesofselecteddescriptors.Inordertoverifytheperformanceofourapproach,weperformexperimentalcomparisonsonPrincetonShapeBenchmarkdatabase.Testresultsshowthatourmethodisapose-oblivious,efficientandrobustnessmethodforeithercompleteorincompletemodels.
出版日期
2013年01月11日（中国期刊网平台首次上网日期，不代表论文的发表时间）
