Signal Processing Image Communication--An efficient macroblock-based diverse and flexible prediction16
上亿文档资料，等你来发现
Signal Processing Image Communication--An efficient macroblock-based diverse and flexible prediction16
Thisarticleappearedinajo；andsharingwithcolleagues；websitesareprohibited.In；encouragedtovisit:；/；Author'spersonalcopy；SignalProcessing:ImageCo；Contentslistsav
ThisarticleappearedinajournalpublishedbyElsevier.Theattachedcopyisfurnishedtotheauthorforinternalnon-commercialresearchandeducationuse,includingforinstructionattheauthorsinstitutionandsharingwithcolleagues.Otheruses,includingreproductionanddistribution,orsellingorlicensingcopies,orpostingtopersonal,institutionalorthirdpartywebsitesareprohibited.Inmostcasesauthorsarepermittedtoposttheirversionofthearticle(e.g.inWordorTexform)totheirpersonalwebsiteorinstitutionalrepository.AuthorsrequiringfurtherinformationregardingElsevier’sarchivingandmanuscriptpoliciesareencouragedtovisit:/copyrightAuthor's personal copySignalProcessing:ImageCommunication25(ContentslistsavailableatScienceDirectSignalProcessing:ImageCommunicationjournalhomepage:/locate/imageAnef?cientmacroblock-baseddiverseand?exiblepredictionmodesselectionforhyperspectralimagescodingFanZhaoa,b,GuizhongLiua,n,XingWangaabSchoolofElectronicandInformationEngineering,Xi’anJiaotongUniversity,Xi’an710049,ChinaDepartmentofInformationScience,Xi’anUniversityofTechnology,Xi’an710048,ChinaarticleinfoArticlehistory:Received18June2008Accepted24July2010Keywords:HyperspectralimagesCompressioncodingH.264/AVCPredictionmodeCorrelationcoef?cientsabstractInthispaper,anef?cientmacroblock-baseddiverseand?exiblepredictionmodesselectionalgorithmisproposedforcodinghyperspectralimages,whichisinspiredbythepredictionschemeofH264/AVC.Here,differentmodesarespeci?edforthecorrespondingmacroblocks(16?16pixelregionsofaband)ofhyperspectralimagesotherthanthewholebandimageusingonlyonereferencebandimageforprediction.Onlythe4?4modeisemployedfortheintra-bandpredictioninviewofthefactthatcorrelationcoef?cientsofpixelsseparatedbynotmorethanfourpixelsinthespatialdomainaregreaterthan0.65atmostcases.Theoptimalreferencebandisdeterminedbythefastreferencebathereafter,thebestpartitionofthecandidatemacroblockintheoptimalreferencebandisfurtherselectedforinter-bandpredictionofthecurrentmacroblock.Thus,thestrongercorrelationinthespectraldirectionorinthespatialdomainisutilizedforthepredictionofthegivenmacroblock.Withacomparablylowmemoryrequirement,thepredictioncodingschemeisproposedtospeeduptheimplementalprocessusingthefastreferencebandselectionalgorithm,theintegerDCTandthequantization,whichjustneedsthemultiplicationandbit-shiftsoperations.SeveralAVIRISimagesareusedtoevaluatetheperformanceofthealgorithm.Theproposedschemeoutperformsthestate-of-the-art3D-basedcompres-sionalgorithmsatlowerrates.Moreover,comparedwiththemethodbyusingallthepredictionmodesofH.264/AVC,about80%encodingtimecanbesavedbyourmethodunderthesameexperimentalcondition.&2010ElsevierB.V.Allrightsreserved.1.IntroductionHyperspectralimagesarerepresentedbyanimagecube,whosebandsareobtainedfromthesamesceneathundredsofcontinuousspectralbandssimultaneously.Astheresultofthedevelopmentofremotesensingtechnol-ogyandtheincreasedinterestinhyperspectralimagesindiverseapplications,thestorageandtransmissionofthethree-dimensional(3D)hyperspectraldatasetshaveCorrespondingauthor.Tel./fax:+.E-mailaddresses:vcu@(F.Zhao),liugz@mail.(G.Liu),wangxing@stu.(X.Wang)./$-seefrontmatter&2010ElsevierB.V.Allrightsreserved.doi:10.1016/j.image.nbecomesigni?cant,whichmakesef?cientcompressionofhyperspectralimagesanactiveresearcharea.Boththespatialandspectralcorrelationsexistinthehyperspectralimages.Howtomakebestuseofthesetwocorrelationsisthekeytoanef?cientcompressionalgo-rithm.Fortheirhighcompressionef?ciencyandexcellentembeddingfeature,severalpromisinglossycompressionalgorithmsbasedontwo-orthree-dimensionalwavelettransformshavebeenproposedrecently,andtheexperi-mentalresultsshowthat,atthelowercompressionbitratios,thebettermethodsaretheoneswhicharebasedonthethree-dimensionalwavelettransform.Themostclassi-cal3Dwaveletvideocodingalgorithmisthe3DSPIHT(3Dsetpartitioninginhierarchicaltrees)proposedin[1].Itisan698F.Zhaoetal./SignalProcessing:ImageCommunication25(extensionoftheoriginal2DSPIHT[2]andhasa3Dsymmetrictreestructure.Limetal.[3]adoptedthe3DSPIHTalgorithmforhyperspectralimagecompression.The2DSPECK(2DSetPartitioningEmbeddedBlock)coderwasproposedin[4],whichutilizedthequadtreepartition-ingandoctavebandpartitioningcodingscheme.Tang,PearlmanandModestino[5,6]extendedandmodi?editto3DSPECKforhyperspectralimagecompression.Thisalgo-rithmadoptedthe3Dwavelettransformfordecorrelationandtheoctreepartitioningtosortthesigni?canceofpixels.Experimentalresultsshowthat3DSPECKisnotonlycomparablewith3DSPIHTincompressionef?ciencybutalsohaslowercomputationalcomplexity.Boththe3DSPIHTandthe3DSPECKalgorithmsuseaconventionalsymmetric3DDWTandtheyarecloseintherate-distortionperformance.Becausethestatisticsofthehyperspectralimagesarenotsymmetricalongthethreedirections(namelythespatial-horizontal,spatial-vertical,andspectraldirections),thecodingperformanceofthesymmetric3DDWTisnotoptimal.Formoreef?cientlyemployingthehighercorrelationinthespectraldirectionthaninthespatialdomainforthehyperspectralimages,someasym-metric3DDWTbasedmethodshavebeenproposedrecently.AnalgorithmAT-3DSPIHT(asymmetrictreebased3DSPIHT)waspresentedin[7]and[8],andanotheralgorithmAT-3DSPECK(asymmetrictransform3DSPECK)wasproposedin[9]forhyperspectralimagescompression.Theseasymmetric3DDWTalgorithmsexhibitedbetterperformancesthanthecorrespondingsymmetric3DDWTones.PartIIofJPEG2000standard[10]makesprovisionformulti-componentimagecompression,whichisagoodchoiceforhyperspectralimagecompression.Ruckeretal.in[11]?rstpresentedthreeJPEG2000codingstrategiesforhyperspectralimages.Ahybridschemewasproposedin[12].Aprincipalcomponentanalysis(PCA)wasdeployedinJPEG2000toprovidespectraldecorrelationaswellasthereductionofthespectraldimensionalityin[25].AlowcomplexityKLThasbeenintegratedintoaJPEG2000part2compliantschemein[26,27].AlthoughthesuperiordecorrelationcapabilitiesareembodiedintheKLT-andPCA-basedschemes,theyhaveafewdrawbacks.Firstthesignal-adaptivetransformmatrixhastobecomputedforeachoftheinputvectors,whichwouldrequirethesolutiontonumericallyintensiveeigenvectorproblems.Moreover,thetransformmatrixhastobetransmittedalongwiththetransformcoef?cients,thuscausingacodingoverhead.Andinthecaseofhighdegreeofnonstationarity,KLTwouldbecomefarfromoptimal.RaoandBhargava[13]proposedaschemewithasimpleblock-basedinter-bandlinearpredictionfollowedbyablock-basedDCT(discretecosinetransform),whichresemblestheoneusedintheMPEGvideocodingstandards.Intheirscheme,two?xedbandswithalowerwavelengthandahigherwavelength,respectively,aredesignatedasthereferencebands.Aftereachofthereferencebandsiscodedwithhigh?delity?rst,thepreviousadjacentbandandthenearestoneofthetwoselectedreferencebandsareusedforabi-directionalpredictiontothecurrentband.Butincaseofalargenumberofbands,thelongdistancebetweenthecurrentbandandthereferencebandwouldresultindistortionincreaseincodingthepredictionerror,especiallywhenthecorrelationbetweenthecurrentbandanditspreviousadjacentbandisalsocomparablylow.In[22],amuchmoreef?cientscanbased3Dimplementationwasproposedforvideocoding,whichcouldeasilyemploythesocalled‘‘slidingwindow’’DWTtechniquealongthespectraldirectionandthuscouldeasilybedeployedwithintheJPEG2000-MCframework.Althoughsuchasliding-windowJPEG2000-MCwouldbealightrequirementonmemoryandnotverycomputa-tionallycomplex,acompromisehastobefoundbetweenthe?lterlengthandthenumberofdecompositionlevelstodetermineabetterfrequencyanalysis.Foranexample,fora3-leveltemporalwavelettransform,thememoryrequirementofthescanbasedDWTsystemisthatof28framesforthe5-tap5/3?lterbanksand48framesforthe9-tap9/7?lterbanksrespectively,andfora4-leveltemporalwavelettransform,thememoryrequire-mentisthatof49framesforthe5/3?lterbanksand87framesforthe9/7?lterbanks,respectively.Suchamemoryrequirementlike87?512?614?16bitsforband-sequentialformat(BSQ)isnearlyequivalenttothememoryrequirement199?614?224?16bitsforband-interleaved-by-lineformat(BIL)inwhichthenumberofthelinesis199,andthiswouldbeaheavyconstraintontheon-boardcompression.Notethat,forfullymakinguseofthespectralcorrelation,thenumberofthebandsandthelinesdemandedformorethanthreedecompositionlevels3DDWTarerelativelylarge,whichwouldun-doubtedlyconstituteheavyburdenofcomputationandstorage.Exploitingthestrongcorrelation,alinearpredictionbetweenbandswasproposedin[23]togreatlyreducethebitraterequiredtocodeimages,inwhichseveralbandorderingsarecomparedandthebestreverseorderingisusedforallthesimulations.However,inthecomplexityofthereorderingthereisO(m2),wheremisthenumberofbandsofthedataset.Itisnotsofeasibleforonboardcompressionduetoboththememoryandprocessingpowerlimitations.Consideringthefactthatthecorrelationsbetweenacurrentmacroblock(a16?16pixelsregionofabandimage)andtheonesinthecorrespondingreferencebandimagesvarywiththepositionofthemacroblock,weadvocateamacroblock-basedratherthanband-basedmulti-bandpredictionfollowedbyamacroblock-basedintegerDCTandCABACarithmeticcodingembodiedin[14,15]forhyperspectralimages.Eachmacroblockofthecurrentbandispredictedusingthesurroundingmacro-blocksinthecurrentbandorusingthereferencemacro-blocksattheidenticalpositioninthereferencebands.Onepurposeofthispaperistopresentafastalgorithmforthepredictionmodeselection,inwhichthestrongercorrelationischosenforpredictioninthespectraldirectionorinthespatialdomain.SincetheH.264/AVCencoderconsiders3intra-framemodesand7inter-framemodesofpredictions,thecomplexityofthecorrespondingmodeselectionisveryhigh.Toreducethecomplexityandtofullyutilizethecharacteristicsofthehyper-spectralimages,weproposetousetheonemode4?4F.Zhaoetal./SignalProcessing:ImageCommunication25(699forintra-bandpredictionandthelimitedpartitionmodes16?16,16?8,8?16and8?8forinter-bandprediction,andtheoptimalreferencebandisdeterminedbythefastreferencebandselectionalgorithm.Duetoafullutiliza-tionofthe?exiblemacroblock-basedintra-bandandinter-bandpredictionsandthehighef?ciencyofthefastreferencebandselectionalgorithm,aswellastheapplicationoftheintegerDCTtransformandthedivi-sion-freequanti?cation,ourlossycompressionalgorithmforhyperspectralimagesisexpectednotonlytoincreasethespeedofthepredictionprocessbutalsotoachieveagoodcodingperformance.Apreliminaryworkhasbeenbrie?yreportedin[24].Thispaperisorganizedasfollows:theschemeofourcodecisproposedinSection2.Section3presentstheexperimentalresults.Finally,Section4concludesthispaper.2.1.Bandclassi?cationThemulti-framepredictiontechnologyhasbeensuccessfullyacceptedintheH.264/AVC,andthencehaslargelyimprovedthecodingef?ciency.Usingbothpastandfuturebandsasreferencesforpredictionofthecurrentbandwouldundoubtedlyimprovethecodingef?ciency.TheDPB(decodedpicturebuffer)isusedtostoretheolddecodedbandsforthesubsequentreferencebaandsomebandsintheDPBarethepastbandsandtheothersarethefuturebands.So,inordertoclassifythepastband,thefutureband,andtheoneswhicharepredictedbytheformertwobands,I-,P-andB-bandareusedtodesignatethetypeofthecorrespondingband.Asimplebandclassi?cationexampleisshowninFig.2(a).‘‘I-band’’isemployedforonlytheintra-bandprediction,thatistosay,eachofitsmacro-blockarepredictedbythespatiallyneighboringones.‘‘P-band’’isappointedforuni-directionalinter-bandpredictioninwhicheachmacroblockcanbepredictedusingitsspatiallyneighboringonesorthecorrespondingreferenceoneinthepastI-orP-band.‘‘B-band’’isdesignatedforbi-directionalinter-bandpredictioninwhicheachmacroblockcanbepredictedusingitsspatiallyneighboringonesorthecombinationofthepastandthefutureband.So,inthisexample,band‘1I’isusedtoformapredictionforband‘4P’,bands‘1I’and‘4P’arebothusedtoformpredictionsforbands‘2B’and‘3B’.Theencodingorderisfurthermodi?edasinFig.2(b).Thebandsinthefrontrankcanpredicttheonesbehind.Thespeci?cdetailsareintroducedatthelatterstage.2.Anef?cientcompressionalgorithmforhyperspectralimagesbasedondiverseand?exiblepredictionmodesTheproposedmacroblock-basedcodecisillustratedinFig.1.Firstly,afterbeingpre-quantized,classi?edtodifferentbandtypesanddividedintonon-overlappingmacroblocks,theinputbandisfedintotheencoder.Secondly,theoptimalpredictionmodeofthecurrentmacroblockisdeterminedbythefastreferencebandselectionalgorithm.Lastly,thepredictionerrorisDCTtransformed,quantizedandCABACencoded.AndallthesefunctionalelementsarethesameasinH.264/AVC.Thesamplevaluesofthedecodedresidualmacroblocksandthecorrespondingreferencemacroblockaresummedtoformthereconstructedmacroblocks,which?nallycon-stitutethedecodedbandimagesstoredinDPB(decodedpicturebuffer),fromwhichthereferencebandandmacroblockarefurtherselectedformulti-bandpredic-tion.Theformalstepsofourcompressionschemearedescribedbelow,followedbydescriptionstothekeycomponentstherein.2.2.FormalstepsOurcompressionschemeconsistsofthefollowingsteps:Step1:Forutilizingthemulti-bandprediction,alltheinputbandimagesareclassi?edasI-,P-,orB-band,whichisexplainedintheprecedingsubsection.Fig.1.Macroblock-basedcodecstructure.700F.Zhaoetal./SignalProcessing:ImageCommunication25(Fig.2.(a)Bandclassi?cationexample,(b)bandencodingorder.Step2:Inordertosimplifythesubsequentprocessing,pre-quantizationisperformedforthe16bitsAVIRISdata,whichisdescribedinSubsection2.3.Step3:Afastpredictionisdoneforeachofthemacroblockunits.??IfthecurrentbandisanI-band,the4?4intra-bandpredictionmodeisutilized.??IfthecurrentbandisaP-band,theoptimalanchorbandB1refischosenfromtheforwardreferencelistfFm,m?1,2,áááMginDPBusingthefastreferencebandselectionalgorithmtobedescribedinSubsection2.5.3.Afterthat,theoptimalpredictionmodeandthepartitionmodeofthegivenmacroblockinP-bandisthendetermined.??IfthecurrentbandisaB-band,theforwardoptimalanchorbandB12refandthebackwardoneBrefarechosenusingthefastreferencebandselectionalgorithmfromtheforwardreferencelistfFm,m?1,2,ááá,MgandthebackwardreferencelistfBn,n?1,2,ááá,Ng,respectively.Afterthat,theoptimalpredictionmodeandthepartitionmodeofthegivenmacroblockinB-bandisdetermined.Step4:Thedifferencebetweenthecurrentmacroblockandthecandidatereferenceoneisthencalculated,integerDCTtransformedandquanti?ed.Step5:Thepredictionmodeandthepredictionerrorofthemacroblockareencodedusingthecontext-basedadaptivebinaryarithmeticcoding(CABAC)[15],andthe?nalbitstreamofabandisgeneratedbycollectingthesub-bitstreamsofthebandtypeaswellasallitsallmacroblocks.2.3.Pre-quantizationOuralgorithmisbasedontheframeworkofH.264with8bitsdata,sothepre-quanti?cationoftheAVIRISdatato8bitsisdoneintheexperiment.Althoughinter-banddataconsistencyismaintainedandalltheproposeddataunitsaretreatedasequallyaspossible,theexpansiontothe16bitsdataprocessingisthetargetoffuturework.Here,Inordertomaintaininter-banddataconsistencyandtreatalltheprocesseddataunitsasequitablyaspossible,thequantizationofaspeci?cbandshoulddependonitsmaximumpixelvalue.max????Xei,j,nT??Let??bethemaximumabsolutevalueofthei,jspeci?clinesetofthegivenbandn.ThenthenumberoftherightshiftbitstothedatasetisgivenbyFig.3.CorrelationmatrixofCupritesceneimage,(a)beforepre-quanti?cationand(b)afterpre-quanti?cation.log2emax????Xei,j,nT????Tà8t1.Bydoingso,lessinformationi,jwouldbelostatthebandwheretheconsistencyisverylowbetweenthecurrentbandandtheadjacentbandsorwhereitspixelvaluesarecomparablysmall.Moreover,moderateamountofinformationlossdoesnotsigni?-cantlyimpairdataexploitationatsuchbandsinwhichmostpixelvaluesareconsiderablylarge.Fig.3showsthecorrelationmatrices[28]oftheCupritesceneimagebeforeandafterpre-quanti?cation.Theredtonesnearthediagonalofthecorrelationmatrixindicatethattheadjacentbandsaretypicallyhighlycorrelated.Theblue包含各类专业文献、各类资格考试、专业论文、中学教育、行业资料、Signal Processing Image Communication--An efficient macroblock-based diverse and flexible prediction16等内容。　This page uses JavaScript to progressively load the article content as a user scrolls.
Screen reader users, click the load entire article button to bypass dynamically loaded article content.
Login via your institution
Remember me
&RIS&(for EndNote, Reference Manager, ProCite)
&RefWorks Direct Export
& Citation Only
& Citation and Abstract
JavaScript is disabled on your browser.
Please enable JavaScript to use all the features on this page.
JavaScript is disabled on your browser.
Please enable JavaScript to use all the features on this page. This page uses JavaScript to progressively load the article content as a user scrolls. Click the View full text link to bypass dynamically loaded article content.
, November 2012, Pages 74–88
Fatigue life prediction for broad-band multiaxial loading with various PSD curve shapes, , , , , , , a Opole University of Technology, Faculty of Mechanical Engineering, Department of Mechanics and Machine Design, ul. Miko?ajczyka 5, 45-271 Opole, Polandb Department of Mechanics, Biomechanics and Mechatronics, Faculty of Mechanical Engineering, Czech Technical University in Prague, Technicka 4, 16607 Prague, Czech Republicc CDM, Institute of Thermomechanics, Academy of Sciences of the Czech Republic, Veleslav&nova 11, 30100 Plzeň, Czech RepublicThe fatigue calculation procedure analysed here applies the power spectral density (PSD) function of the equivalent stress together with the known spectral method for estimating the probability density function of stress amplitudes included in random loading. Here, the narrow-band approximation, Wirsching&Light, Benasciutti&Tovo and Dirlik models are used, together with the SWT parameter. The prediction capability of these four methods was verified on a set of 107 tests results obtained under random axial, torsion and combined axial and torsion loading applied to a tubular specimen with a one-sided hole. Several PSD shapes and combinations of loading were applied. It is shown that the results for fatigue life calculated using the Benasciutti&Tovo and Dirlik methods are well correlated with the results of experiments under this type of loading.Highlights? Unique multiaxial fatigue tests results (combined axial and torsion random loading). ? Random loading realized with different shapes of PSDs. ? Damage maps for indicating the place of maximum fatigue damage. ? Simple and fast fatigue damage assessment based on spectral method.KeywordsRandom loading; Power spectral density; Fatigue damage; Narrow-band approximation; Broad-band spectral models
Cookies are used by this site. To decline or learn more, visit our
No articles found.
This article has not been cited.
No articles found. 上传我的文档
 下载
 收藏
该文档贡献者很忙，什么也没留下。
 下载此文档
正在努力加载中...
LNCS 3767 - A Cross-Resolution Leaky Prediction Scheme for In-Band
下载积分：500
内容提示：LNCS 3767 - A Cross-Resolution Leaky Prediction Scheme for In-Band
文档格式：PDF|
浏览次数：2|
上传日期： 13:15:13|
文档星级：
该用户还上传了这些文档
LNCS 3767 - A Cross-Resolution Leaky Prediction Sc
官方公共微信