An affine invariant discriminate analysis with canonical correlation analysis

Canonicalcorrelationanalysis(CCA)isinvariantwithregardtoaffinetransformation,butitcannotbe directlyappliedtoaffineinvariantpatternrecognition.Thereasonmainlyliesinthatmanyexisting CCA-basedschemesrepresentthepatternbymatrix-to-vectormethod,asaresult,thestructureand spatialinformationoftheoriginalpatternisdiscarded.Inthispaper,anaffineinvariantdiscriminate analysis(AIDA)methodisdevelopedforpatternrecognition.Dislikethematrix-to-vectorrepresenta- tion, anobjectisfirstconvertedtoaprojectionmatrixbycentralprojectiontransform(CPT).Aftera point matchingprocess,CCAisperformedtoprojectionmatricesoftheobjectandthemodel,andtwo vectorswillbederived.Therefore,theobjectisclassifiedtoamodelbythesmallestdistancebetween the obtainedvectors.Comparisonsofexperimentalresultsaregivenwithrespecttosomeexisting methods,whichdemonstratetheeffectivenessoftheproposedAIDAmethod.