基于感知不變性的目標特征提取技術研究.pdf

1、匾中國石油大學碩士研究生學位論文華東(申請工學碩士學位)基于感知不變性的目標特征提取技術研究學科專業(yè)：培養(yǎng)方向：碩士生：指導教師：機械設計及理論機械設計方法與理論劉敬劉衍聰(教授)入學日期：2005年9月論文完成日期：2007年4月STUDYONFEATUREEXTRACTIONTECHNIQUEBASEDONTHEPERCEPTIVEINVARIABILITYLIUJing(MechanicalDesignandTheory)Dire

2、ctedbYProfessorLIUYaneongAbstractFeatureextractionistheoneoftheimportanttechniquesinpaaemrecognitionHowtoextraetthesteadyandreliablefeatureisthekeysteptoaccomplishtherecognitionImageswillbeaffectedbythedisplacement，scale

3、，rotationeomplicatedvaryingilluminationThisthesismostlyconsiderstheilluminationinvariantinthefeatureextractiontechnique，andgivesattentiontothedisplacementscale，rotationinvariabilityOuraimistoextractthefeatureMtlldisplace

4、ment，scale，rotationilluminationinvariabilityThethesistheoreticallystudiesthefeatureextractiontechniquebasedontheperceptiveinvariantsApplyandexpandHumomeminvariantsbytheexperimentthengainthedisplacemem，scaie，rotationinvar

5、iability111isthesissumsupthemethodSonhowtoeliminatetheeffectsfromthecomplicatedvaryingilluminationAmo(1el—basedLocalNormalizationTechniqueandRelativeGradientareadoredforilluminationinvariance11鹼LNalgorithmworksbyillumina

6、tionmodeltoestimatetheeffects，andsimulatesilluminationeffectswiththemultiplicativenoiseandtheadditivenoiseTheRelativeGradientalgorithmimprovesthe可adiem，extractstheinnercharacterastheilluminationinvariantAdopttheYaleBface