简介:在核范数鲁棒主成分分析的基础上,利用加权Schatten-p范数和l2,1范数重新构造鲁棒的主成分分析问题,使得原始稀疏正则化、秩最小化问题得到了较好的非凸逼近.建立一个新的基于加权Schatten-p范数和l2,1范数的鲁棒主成分分析(WLSRPCA)模型,并使用增广拉格朗日乘子法进行求解.在图像去噪的实验中,WLSRPCA模型去噪效果比鲁棒主成分分析模型更好.
简介:1.IntroductionInthispaperweconsiderCauchyproblemforaclassofnonhomogeneousNavier-Stokesequationsintheinfinitecylinderwith.Givensatisfyinginthedistributionsensediv,weseekasolutionvectorandapressurefunctionP(t,x)suchthatwhereisanonlinearvector-valuedfun...
简介:Inthispaperwewillshowthatifanapproximationprocess{Ln}n∈Nisshapepreservingrelativetotheconeofallk-timesdifferentiablefunctionswithnon-negativek-thderivativeon[0,1],andtheoperatorsLnareassumedtobeoffiniterankn,thentheorderofconvergenceofDkLnftoDkfcannotbebetterthann2evenforthefunctionsxk,xk+1,xk+2onanysubsetof[0,1]withpositivemeasure.Takingintoaccountthisfact,wewillbeabletofindsomeasymptoticestimatesoflinearrelativen-widthofsetsofdifferentiablefunctionsinthespaceLp[0,1],p∈N.
简介:Motiondeblurringisabasicprobleminthefieldofimageprocessingandanalysis.Thispaperproposesanewmethodofsingleimageblinddeblurringwhichcanbesignificanttokernelestimationandnon-blinddeconvolution.Experimentsshowthatthedetailsoftheimagedestroythestructureofthekernel,especiallywhentheblurkernelislarge.SoweextracttheimagestructurewithsalientedgesbythemethodbasedonRTV.Inaddition,thetraditionalmethodformotionblurkernelestimationbasedonsparsepriorsisconducivetogainasparseblurkernel.Butthesepriorsdonotensurethecontinuityofblurkernelandsometimesinducenoisyestimatedresults.ThereforeweproposethekernelrefinementmethodbasedonL0toovercometheaboveshortcomings.Intermsofnon-blinddeconvolutionweadopttheL1/L2regularizationterm.Comparedwiththetraditionalmethod,themethodbasedonL1/L2normhasbetteradaptabilitytoimagestructure,andtheconstructedenergyfunctionalcanbetterdescribethesharpimage.Forthismodel,aneffectivealgorithmispresentedbasedonalternatingminimizationalgorithm.
简介:LetL~2([0,1],x)bethespaceoftherealvalued,measurable,squaresummablefunctionson[0,1]withweightx,andlet■_nbethesubspaceofL~2([0,1],x)definedbyalinearcombinationofJ_0(μ_kx),whereJ_0istheBesselfunctionoforder0and{μ_k}isthestrictlyincreasingsequenceofallpositivezerosofJ_0.Forf∈L~2([0,1],x),letE(f,■_n)betheerrorofthebestL~2([0,1],x),i.e.,approximationoffbyelementsof■_n.Theshiftoperatoroffatpointx∈[0,1]withstept∈[0,1]isdefinedbyT(t)f(x)=(1/π)∫_0~πf((x~2+t~2-2xtcosθ)~(1/2))dθ.Thedifferences(1-T(t))~(r/2)f=∑_(j=0)~∞(-1)~j(_j~(r/2))T~j(t)foforderr∈(0,∞)andtheL~2([0,1],x)-modulusofcontinuityω_r(f,τ)=sup{||(I-T(t))~(r/2)f||:0≤t≤τ}oforderraredefinedinthestandardway,whereT~0(t)=Iistheidentityoperator.Inthispaper,weestablishthesharpJacksoninequalitybetweenE(f,■_n)andω_r(f,τ)forsomecasesofrandτ.Moreprecisely,wewillfindthesmallestconstant■_n(τ,r)whichdependsonlyonn,r,andτ,suchthattheinequalityE(f,■_n)≤■_n(τ,r)ω_r(f,τ)isvalid.
简介:Thepurposeofthepresentpaperistoevaluatetheerroroftheapproximationofthefunc-tionfL1[0,1]byKantorovich-BernsteinpolynomialsinLp-metric(0
简介:Inthispaper,simultaneousuniformapproximationandmeanconvergenceofquasi-HermiteinterpolationanditsderivativebasedonthezerosofJacobipolynomialsareconsideredseparately.Thedegreesofthecorrespondingapproximationsarerespectivelygivenalso.Someknownresultsareimprovedaudextended.