We focus on the minimization of regularized objective functions using the popular half-quadratic approach introduced by Geman and Reynolds in 1992. We show that whenever applicable, this approach is equivalent to the very classical gradient linearization a
Half-QuadraticMinimizationofRegularizedObjectivesastheFixedPointIterationforthe
LinearizedGradient
MilaNikolovaandRaymondChan
Abstract
Wefocusontheminimizationofregularizedobjectivefunctionsusingthepopularhalf-quadratic
approachintroducedbyGemanandReynoldsin1992.Weshowthatwheneverapplicable,thisapproach
isequivalenttotheveryclassicalgradientlinearizationapproach,knownalsoasthe xedpointiteration.
Keywords:half-quadraticregularization,inverseproblems,gradientlinearization, xedpointiteration,optimization,signalandimagerestoration,variationalmethods.
I.INTRODUCTION
Letdatay∈Rqbeobtainedfromanoriginalunknownimageorsignalx viay=Ax +noisewhereA∈Rq×pisalineartransform.Typicalapplicationsaredenoising,deblurring,imagereconstructionintomographyandotherinverseproblems[1],[2],[3].Thesought-afterx ∈Rpisde nedastheminimizerofanobjectivefunctionJ:Rp→Roftheform
J(x)= Ax y 2+βΦ(x),
r Φ(x)= ( Gix ),
i=1(1)(2)
whereGi:Rp→Rs,s≥1fori=1,...,r,arelinearoperators, . isthe 2norm, :R+→Riscalledapotentialfunction(PF)andβ>0isaparameter.Typically1,eithers=1andoperatorsGiCentredeMath´ematiquesetdeLeursApplications(CNRS-UMR8536),ENSdeCachan,61avenueduPresidentWilson,94235CachanCedex,France(nikolova@cmla.ens-cachan.fr).
DepartmentofMathematics,TheChineseUniversityofHongKong,Shatin,HongKong(rchan@math.cuhk.edu.hk).ThisworkwassupportedbyHKRGCGrantCUHK400503andCUHKDAG2060257.
热门文档