简介:Thispaperpresentssomeresultsoftherelationbetweenwavelettransformandfractaltransform.Thewavelettransformoftheattractoroffractaltransformpossesestranslationalandscaleinvariance.Sowespeedthefractalimageencodingbytestingtheinvarianceofthewavelettransformappropriateforimageencoding.Theclassficationschemeofrangeblocksbywavelettransformisgiveninthispaper.
简介:Inimagereconstructionalgorithms,thechoicesoffilterfunctionsandinterpolatingfunctionsareveryimportantforthecomputationalspeedandthequalityoftheimagereconstructed,especially,forfan-beamgeometry,theoccurrenceofthesingularintegraloperatormayleadtosomegreatoscillationscomparedtotheoriginalimage.Inthispaperwewillgiveadirectconvolu-tionalgorithmwhichneedsnotthecomplexcomputationsoccuringintheFouriertransform,thenusingacircleintegralweobtainastablecomputationalprogram.Differentfromallotherpreviouswindowfunctionsusedbymanypioneerresearchers,inouralgorithmwechooseawindowfunctionsimilartoGabor'swindowfunctione-x^2/2,whichcanberegardedastheapproximationtotheinverseFouriertransformofalocallyintegrablefrequencyfunction.AlsowepointoutthatsuchreconstructionalgorithmprocedurescanbeusedtodealwiththeSPECTprojectiondatawithconstantattenuation.
简介:Anewsmoothingmethodisproposed.Thesmoothingprocessadaptstoimagecharacteristicsandisgoodatpreservinglocalimagestructures.Moreimportantly,inthetheoryundertheconditionsweakerthanthoseintheoriginalKacanovmethodanapproximalsequenceofsolutionstothevariationalproblemscanbeconstructedandtheglobalconvergencecanbeproved.AndtheconditionsinthepapersofSchn6rr(1994)andHeers,etal(2001)arediscussed.Numericalsolutionsofthemodelaregiven.
简介:Inthispaper,weproposeadiscrepancyrule-basedmethodtoautomaticallychoosetheregularizationparametersfortotalvariationimagerestorationproblems.Theregularizationparametersareadjusteddynamicallyineachiteration.Numericalresultsareshowntoillustratetheperformanceoftheproposedmethod.
简介:Weproposeanewalgorithmforthetotalvariationbasedonimagedenoisingproblem.ThesplitBregmanmethodisusedtoconvertanunconstrainedminimizationdenoisingproblemtoalinearsystemintheouteriteration.Analgebraicmulti-gridmethodisappliedtosolvethelinearsystemintheinneriteration.Furthermore,Krylovsubspaceaccelerationisadoptedtoimproveconvergenceintheouteriteration.Numericalexperimentsdemonstratethatthisalgorithmisefficientevenforimageswithlargesignal-to-noiseratio.
简介:Inthispaper,westudyanoperatorswhichmapseveryn-by-nsymmetricmatrixA,toamatrixs(A_n)thatminimizes||B_n-A_n||FoverthesetofallmatricesB_n,thatcanbediagonalizedbythesinetransform.Thematrixs(A_n),calledtheoptimalsinetransformpreconditioner,isdefinedforanyn-by-nsymmetricmatricesA_n.Thecostofconstructings(A_n)isthesameasthatofoptimalcirculantpreconditionerc(A_n)whichisdefinedin[8],Thes(A_n)hasbeenprovedin[6]tobeagoodpreconditionerinsolvingsymmetricToeplitzsystemswiththepreconditionedconjugategradient(PCG)method.Inthispaper,wediscussthealgebraicandgeometricpropertiesoftheoperators,andcomputeitsoperatornormsinBanachspacesofsymmetricmatrices.Somenumericaltestsandanapplicationinimagerestorationarealsogiven.
简介:Imagerestorationisoftensolvedbyminimizinganenergyfunctionconsistingofadata-fidelitytermandaregularizationterm.Aregularizedconvextermcanusuallypreservetheimageedgeswellintherestoredimage.Inthispaper,weconsideraclassofconvexandedge-preservingregularizationfunctions,I.e.,multiplicativehalf-quadraticregularizations,andweusetheNewtonmethodtosolvethecorrespondinglyreducedsystemsofnonlinearequations.AteachNewtoniterate,thepreconditionedconjugategradientmethod,incorporatedwithaconstraintpreconditioner,isemployedtosolvethestructuredNewtonequationthathasasymmetricpositivedefinitecoefficientmatrix.Theigenvalueboundsofthepreconditionedmatrixaredeliberatelyderived,whichcanbeusedtoestimatetheconvergencespeedofthepreconditionedconjugategradientmethod.Weuseexperimentalresultstodemonstratethatthisnewapproachisefficient,andtheeffectofimagerestorationisr0easonablywell.