简介:IntheH.263videocodecrelatedsystems,motionestimationandDiscreteCosineTransform(DCT)havethemostcomputationalrequirements.Inordertoreducecomplexityoftheencodertodedicatemoreresourcestootherfunctions,accordingtothestudyofexistingmethods,anImprovedAllZeroBlockFinding(IAZBF)methodbasedonthestatisticcharacteristicsofDCTcoefficientsisproposed.Comparedwithexistingmethods,IAZBFimprovesthedetectingefficiencybyabout50%withoutimportingtoomuchextracomputationrequirement.Beingcomputedwithadditionsandshiftsinsteadofcomplicatedmultiplications,IAZBFisoflowcomputationcomplexity,especiallyforlow-endprocessors.Inaddition,IAZBFupholdspicturefidelityandremainscompatiblewiththeH.263bitstreamstandard.
简介:Tocompresshyperspectralimages,alowcomplexitydiscretecosinetransform(DCT)-baseddistributedsourcecoding(DSC)schemewithGraycodeisproposed.UnlikemostoftheexistingDSCschemes,whichutilizetransforminspatialdomain,theproposedalgorithmappliestransforminspectraldomain.Set-partitioning-basedapproachisappliedtoreorganizeDCTcoefficientsintowaveletliketreestructureandextractthesign,refinement,andsignificancebitplanes.TheextractedrefinementbitsareGrayencoded.Becauseofthedependencyalongthelinedimensionofhyperspectralimages,lowdensityparitycheck-(LDPC)-basedSlepian-WolfcoderisadoptedtoimplementtheDSCstrategy.Experimentalresultsonairbornevisible/infraredimagingspectrometer(AVIRIS)datasetshowthattheproposedparadigmachievesupto6dBimprovementoverDSC-basedcoderswhichapplytransforminspatialdomain,withsignificantlyreducedcomputationalcomplexityandmemorystorage.
简介:Inthispaper,twospeechenhancementsystemswithsupergaussianspeechmodelingarepresented.ThecleanspeechcomponentsareestimatedbyMinimum-Mean-Square-Error(MMSE)es-timatorundertheassumptionthattheDCTcoefficientsofcleanspeecharemodeledbyaLaplacianoraGammadistributionandtheDCTcoefficientsofthenoiseareGaussiandistributed.Then,MMSEestimatorsunderspeechpresenceuncertaintyarederived.Furthermore,theproperestimatorsofthespeechstatisticalparametersareproposed.ThespeechLaplacianfactorisestimatedbyanewdeci-sion-directedmethod.ThesimulationresultsshowthattheproposedalgorithmyieldslessresidualnoiseandbetterspeechqualitythantheGaussianbasedspeechenhancementalgorithmsproposedinrecentyears.
简介:BasedontheQuikSCATdata,thefeaturesofsurfacewinddistributionofthetyphoonVongfonglandfallprocessareanalyzed.WehavealsostudiedthevariancespectralconfigurationofthesurfacewindfieldusingDCT(DiscreteCosineTransform).Theconclusionsareasfollows:Thenear-surfacewindfieldishighlyasymmetric;thevariancecomponentsofasymmetricsurfacewindfielddependmainlyontheairflowdirectionofwavenumber1and2.Whenthetyphoonmoveswest,therearetwowavespectralcentersliningupinthezonaldirection,mainlytheairflowfromzonalwavenumber2andmeridionalwavenumber2;whenitmovesnorth,therearetwowavespectralcentersinameridionalarray,mainlytheairflowfromzonalwavenumber1andmeridionalwavenumber2.Theairflowforwavenumber1mainlycontributestothevarianceofthetangentialwindwhilethatforwavenumber2tothevarianceoftheradialwind.Theasymmetricaldistributionchangeswiththelarge-scaleenvironmentandself-rotatingcirculationaroundthetyphoon.Whenitapproachedland,theassociatedgaleappearsinfrontportionintheadvancingdirectionofthestorm.ItisineffectsimilartothemodelofChenLian-shoufortyphoon-relatedgales-NNWontheleftfrontportionandSEontherightfrontportion.