简介:Motiondeblurringisabasicprobleminthefieldofimageprocessingandanalysis.Thispaperproposesanewmethodofsingleimageblinddeblurringwhichcanbesignificanttokernelestimationandnon-blinddeconvolution.Experimentsshowthatthedetailsoftheimagedestroythestructureofthekernel,especiallywhentheblurkernelislarge.SoweextracttheimagestructurewithsalientedgesbythemethodbasedonRTV.Inaddition,thetraditionalmethodformotionblurkernelestimationbasedonsparsepriorsisconducivetogainasparseblurkernel.Butthesepriorsdonotensurethecontinuityofblurkernelandsometimesinducenoisyestimatedresults.ThereforeweproposethekernelrefinementmethodbasedonL0toovercometheaboveshortcomings.Intermsofnon-blinddeconvolutionweadopttheL1/L2regularizationterm.Comparedwiththetraditionalmethod,themethodbasedonL1/L2normhasbetteradaptabilitytoimagestructure,andtheconstructedenergyfunctionalcanbetterdescribethesharpimage.Forthismodel,aneffectivealgorithmispresentedbasedonalternatingminimizationalgorithm.
简介:Theaimofthispaperistostudythepracticalф0-stabilityinprobability(Pф0SiP)andpracticalф0-stabilityinpthmean(Pф0SpM)ofswitchedstochasticnonlinearsystems.Sufficientconditionsonsuchpracticalpropertiesareobtainedbyusingthecomparisonprincipleandthecone-valuedLyapunovfunctionmethods.Also,basedonanextendedcomparisonprinciple,aperturbationtheoryofswitchedstochasticsystemsisgiven.