摘要
Inpracticalapplications,weoftenhavetodealwithhigh-orderdata,forexample,agrayscaleimageandavideoclipareintrinsicallya2nd-ordertensoranda3rd-ordertensor,respectively.Inordertosatistythesehigh-orderdata,itisconventionaltovectorizethesedatainadvance,whichoftendestroystheintrinsicstructuresofthedataandincludesthecurseofdimensionality.Forthisreason,weconsidertheproblemofhigh-orderdatarepresentationandclassification,andproposeatensorbasedfisherdiscriminantanalysis(FDA),whichisageneralizedversionofFDA,namedasGFDA.ExperimentalresultsshowourGFDAoutperformstheexistingmethods,suchasthe2-directional2-dimensionalprincipalcomponentanalysis((2D)2PCA),2-directional2-dimensionallineardiscriminantanalysis((2D)2LDA),andmultilineardiscriminantanalysis(MDA),inhigh-orderdataclassificationunderalowercompressionratio.
出版日期
2014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)