简介:Motiondeblurringisabasicprobleminthefieldofimageprocessingandanalysis.Thispaperproposesanewmethodofsingleimageblinddeblurringwhichcanbesignificanttokernelestimationandnon-blinddeconvolution.Experimentsshowthatthedetailsoftheimagedestroythestructureofthekernel,especiallywhentheblurkernelislarge.SoweextracttheimagestructurewithsalientedgesbythemethodbasedonRTV.Inaddition,thetraditionalmethodformotionblurkernelestimationbasedonsparsepriorsisconducivetogainasparseblurkernel.Butthesepriorsdonotensurethecontinuityofblurkernelandsometimesinducenoisyestimatedresults.ThereforeweproposethekernelrefinementmethodbasedonL0toovercometheaboveshortcomings.Intermsofnon-blinddeconvolutionweadopttheL1/L2regularizationterm.Comparedwiththetraditionalmethod,themethodbasedonL1/L2normhasbetteradaptabilitytoimagestructure,andtheconstructedenergyfunctionalcanbetterdescribethesharpimage.Forthismodel,aneffectivealgorithmispresentedbasedonalternatingminimizationalgorithm.
简介:Inthispaper,thesmoothconnectionbetweentwoB-splinesurfacesisdiscussed.First,abriefproofofsomesimplesufficientconditionsofG~0andG~1continuityisgiven.Onthisbasis,anovelmethodforG~0orG~1connectionbetweentwoadjacentB-splinesurfacesispresented.Areparameterizationstepisfirstlytakenforoneofthesurfacessuchthattheyhavethesameparameterizationinvdirection,then,adjusttheirboundarycontrolverticestomakethemG~0orG~1connected.TheG~1connectionparameterisdeterminedbyanoptimizationproblem.Comparedwiththeexistedmethods,ourmethodissimpleandeasytobeusedinpractice.