正則超分辨率圖像復(fù)原算法研究.pdf

1、西北工業(yè)大學(xué)碩士學(xué)位論文正則超分辨率圖像復(fù)原算法研究姓名：張正賢申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算數(shù)學(xué)指導(dǎo)教師：蔣大為20060301胡北r業(yè)人學(xué)碩十學(xué)位論文摘要AbstractSuperresolutionimagerestorationisillposedinverseproblem；itshouldberesolvedbyregularizationmethodSuper—resolutionimagerestorationinvolv

2、esinmulti—flamelowresolutionimageandaflamehighresolutionimage，soitscomputationalcomplexityislargeWhile，theestimationofdegrademodelanditsparameterandthesub—pixelmotionestimationareliabletocreateerrorTheadditionalnoisetype

3、indegradeprocessisdifficulttodetermineConsequentlyweneedtofindfastalgorithmandweexpectithaverobustnesstoaboveerrorsIntraditionalapproach，peopleuse，2一normtomeasuredataapproximationitemandregularizationiteminregularizedima

4、gerestorationandsuperrestorationinspatialdomainSFarsiuetalpointoutuse，一normtomeasuredataapproximationitemandregularizationitemandprovideusarobustsuperresolutionmethod，whilewecan’tobtainsobustnessby，2一normTheydidn’tdiscus

5、stheuniquenessofminimumpointofobjectivefunctionalintheoryBesidesthisproblem，theconvergenceofsteepestdescendalgorithmusedtofindtheminimumpointofobjectivefunctionalandhowtochoosestepsizeparameterandregularizationparametera

6、ren’tdiscussedThispapergivefurtherdiscussandobtainsomeimportantconclusionWediscusstheuniquenessofminimumpointofobjectivefunctionalobtainedbyusing／。norm(1≤P≤2)tomeasuredataapproximationitemandregularizationitemintheoryIfl

7、P≤2，theobjectivefunctionalhasuniqueminimumpoint；ifp=1wecan’tdeterminetheuniquenesseasilyFormakingupthisbug，weuse，。一norm，Pisalittlelargerthan1，andobtainsimilarimagerestorationeffectThispapergivesamethodtochooseregularizat

8、ionparameterDiscusstheperformanceofsteepestdescendalgorithm，introducetheideathatsetthestepsizeparameterinphaseinordertoreduceiterativetimeandimprovetheeffectAdoptRobertscrossgradientoperatorwegetsuperresolutionalgorithmw

9、ithlowercomputationalcomplexityandgetbettereffectthanbilateraltotalvariationoperatoratsomecaseWediscovertheregularizationmethodwith1】一normismorerobusttotheblurestimationerrorthanthatwith，2normTherefore，weapplyittoblindsu