Multiscale Signal Processing and Shape Analysis

《Multiscale Signal Processing and Shape Analysis》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、ABSTRACTHE,YUN.MultiscaleSignalProcessingandShapeAnalysisforanInverseSARImagingSystem.(UnderthedirectionofProf.HamidKrim.)Thegreatchallengeinsignalprocessingistodevisecomputationallyefficientandstatisti-callyoptimalalgorithmsforestimatingsignalsfromnoisybackgroundandunderstandingtheircontents.Thisth

2、esistreatstheproblemofmultiscalesignalprocessingandshapeanalysisforanInverseSyntheticApertureRadar(ISAR)imagingsystem.Toaddresssomeofthelimita-tionsofconventionaltechniquesinradarimageprocessing,aninformationtheoreticapproachfortargetmotionestimationisfirstproposed.Awaveletbasedmultiscalemethodforsh

3、apeenhancementissubsequentlyderivedandfollowedbyaregressiontechniqueforshaperecog-nition.Buildingonentropy-baseddivergencemeasureswhichhaveshownpromisingresultsinmanyareasofengineeringandimageprocessing,weintroduceinthisthesisanewgeneralizeddivergencemeasure,namelytheJensen-R´enyidivergence.Uponest

4、ablishingitspropertiessuchasconvexityanditsupperboundetc.,weapplyittoimageregistrationforISARfocusingaswellasrelatedproblemsindatafusion.Attemptingtoextendcurrentapproachestosignalestimationinawaveletframework,whichhavegenerallyreliedontheassumptionofnormallydistributedperturbations,wepro-poseanove

5、lnon-linearfilteringtechnique,asapre-processingstepfortheshapesobtainedfromanISARimagingsystem.Thekeyideaistoprojectanoisyshapeontoawaveletdomainandtosuppresswaveletcoefficientsbyamaskderivedfromcurvatureextremainitsscalespacerepresentation.Forapiecewisesmoothsignal,itcanbeshownthatfilteringbythiscurv

6、aturemaskisequivalenttopreservingthesignalpointwiseH¨olderexponentsatthesingularpoints,andtoliftingitssmoothnessatalltheremainingpoints.Toidentifyashapeindependentlyofitsregistrationinformation,weproposematchingtwoconfigurationsbyregression,usingnotationsofgeneralshapespacesandprocrusteandis-tances.

7、Inparticular,westudythegeneralizedmatchingbyestimatingmeanshapesintwodimensions.Simulationresultsshowthatmatchingbywayofameanshapeismorerobustthanmatchingtargetshapesdirectly.MULTISCALESIGNALPROCESSINGANDSH