《电子信息类专业英语 》课件-第10章

Unit10ImagesandTelevisionsPassageADigitalImageFundamentalsPassageBCompression/DecompressionTechniquesPassageCTelevision

PassageADigitalImageFundamentals

1.DigitalImageResolution

Adigitalimageismadeupofmanyrowsandcolumnsofpixels.Forgrayscaleimages,eachpixelisassignedanumberthatrepresentsthegrayshadeassignedtothatpixel.Thelargerthenumberofpixelsinanimage,andthelargerthenumberofavailablegrayscalelevels,thebettertheresolutionoftheimage.Figure17.1isan17-bitimages,with217＝256possiblegrayscalelevels.Thenumberofrowandcolumnare303×228.Figure10.1An303×228×17DigitalImage

2.Histograms

Thegrayscalespresentinadigitalimagecanbesummarizedbyitshistogram(seeFigure17.2)．Thehistogramreportsthenumberofpixelsforeachgrayscalelevelpresentintheimage,asabargraph.Whenanimageusesonlyasmallportionoftheavailablegrayscalelevels,histogramequalizationcanbeusedtospreadouttheusageofgrayscalelevelsovertheentireavailablerange［1］．Thisprocedurere-assignsgrayscalelevelssothatimagecontrastisimproved.

Figure10.2AHistogram

3.AdditionandSubtractionofImages

Digitalimagescanbeaddedandsubtractedpixel-by-pixel.Addingtwoimagescancombinetwosetsofobjectsintoasingleimage.Moreover,severalnoisyimagesofthesamescenecanbeaveragedtogethertoreducetheeffectofnoise.Imagesubtraction,ontheotherhand,canbeusedtoremoveanunwantedbackgroundfromanimage.Subtractionoftwotime-lapsedphotographswillshowwheremotionhasoccurredbetweenthetwo.

Whentwoimagesareaddedorsubtracted,theresultantmatrixwillfrequentlycontainillegalgrayscalevalues.Forexample,whenapixelinone17-bitimagehasthegrayscalelevel129andthecorrespondingpixelinasecond17-bitimagehasthegrayscalelevel201,thesumpixelis129+201＝330.Thisisoutsidethelegalrangeforan17-bitimage,whichmayonlycontaingrayscalelevelsbetween0and255.Whenthesametwoimagesaresubtracted,thedifferencepixelis129-201=-72,againavalueoutsidethelegalrange.

Forthesereasons,scalingfollowsmostimagearithmetic.Scalingtotherange［0,GSLmax］maybeaccomplishedasfollows:

4.WarpingandMorphing

Warpingandmorphingaredigitalimagetechniquesthatarefindingapplicationnotonlyinentertainmentbutalsoinmedicalimaging.Warpingstretchesorre-shapesanobjectinanimage,whilemorphingtransformsoneimageintoanother.Thesetransformationsmaybeaccomplishedbymarkingcontrolpoints,controllines,ortrianglesinasourceimageandchoosingtheirnewpositionsinadestinationimage.

Thetransitionbetweensourceanddestinationimagesisthenaccomplishedbysmoothlytransformingnotonlythecontrolelementlocations,butalsotheircolors.Thelocationsandcolorsofpixelsnotexplicitlymarkedascontrolelementsaredeterminedbythelocationsandcolorsofthecontrolelementsthatarenearest.

5.ImageFiltering

Digitalimagescanbefilteredusingtwo-dimensionalconvolutionwithaconvolutionkernel.WhenanN×NimageisfilteredbyanM×Mconvolutionkernel,(M-1/2)rowsandcolumnsoneachsideoftheimagearelosttoboundaryeffects.Lowpassfiltersblurimages,highpassfiltersemphasizesharpchangesingrayscalelevel,andedgefilters

locateedgesinanimage.［2］

6.DilationandErosion

Dilationaddsalayerofpixelstoallobjectsinanimage.Erosionremovesonelayerofpixelsfromallobjects.Whendilationisfollowedbyerosion,gapsinbrokenboundaries

identifiedthroughedgedetectioncanbefilledin.Conversely,whenerosionisfollowedbydilation,spotsofnoiseinanimageareremoved.

Successfullydetectingtheedgesinanimageisthefirststeptowardsconfidentidentificationofobjectboundariesandthenobjectsrecognition.Fromboundaryinformation,shapecharacteristicslikeperimeterandareacanbecalculated,whichcanbeusedtoclassifyanobject.

7.ImageSpectra

Two-dimensionalFFTsareusedtoanalyzethespectraofdigitalimages.Justasintheone-dimensionalcase,atwo-dimensionalspectrumcomprisesamagnitudespectrumandaphasespectrum.Thephasespectrumcarriesthebestinformationaboutthelocationsoftheobjectsintheimage.［3］Whenallmagnitudesaresettoone,thephasesalonestillshowafacsimileoftheoriginalimage.Whenallphasesaresettozero,themagnitudesaloneshownotraceofit.

ImagespectraformthebasisforbothCT(computedtomography)andMRI(magneticresonanceimaging)scandisplays.CTscansareX-raystakeninmanydirectionsinasingleplaneofanobject.［4］

MRIscansdependinsteadonthemagneticpropertiesofanobjectplacedinavaryingmagneticfield.Bothtypesofscanspermitnon-invasiveinvestigationsofthree-dimensionalobjects.

8.Imagecompression

InpartduetotheInternet,digitalimagesaretransmittedfromplacetoplacemore

oftenthanever.Tosavetimeandbandwidth(space),bothimagesandotherfilesareoftencompressedbeforebeingtransmitted.Losslesscompressionmeansthatafileiscompactedwithoutlosinganyinformation,sothatthereconstructedfileisidenticaltotheoriginal.［5］

Lossycompressionmeansthatsomeinformationfromtheoriginalfileisirretrievablylost,butgenerallythereconstructedfileisextremelyclosetotheoriginal.Thecompression

ratioistheratiooftheoriginalfilesizetothecompressedfilesize.

Onesimplecompressionschemeisrun-lengthencoding,whichcodesmorethanthreerepetitionsofanumberasthreecopiesofthenumberfollowedbyacountoftheadditionalcopiesneeded.AnothercompressionschemeisHuffmanencoding,whichusesshortercodestorepresentthemostcommonsignalelements,andlongercodestorepresenttheleastcommonsignalelements.

JPEG,anextremelycommonimagecompressionscheme,usesthediscretecosinetransform(DCT)toconcentratemostoftheinformationaboutan17×17sub-blockofanimageintoafewsignificantcoefficients.［6］

Itthenusesbothrun-lengthencodingandHuffmanencodingtoprovidefurthercompression.

Notes

［1］Whenanimageusesonlyasmallportionoftheavailablegrayscalelevels,histogramequalizationcanbeusedtospreadouttheusageofgrayscalelevelsovertheentireavailablerange.

当一幅图像只使用了可用灰度级的一小部分时,可以使用直方图均衡的方法将灰度级的使用扩展到整个可用的范围。

［2］Lowpassfiltersblurimages,highpassfiltersemphasizesharpchangesingrayscalelevel,andedgefilterslocateedgesinanimage．

低通滤波器使图像变得模糊,而高通滤波器突出了图像的灰度锐变,边缘滤波器则对图像边缘进行定位。

［3］Thephasespectrumcarriesthebestinformationaboutthelocationsoftheobjectsintheimage．

相位谱携带着图像中目标位置的信息。

［4］CTscansareX-raystakeninmanydirectionsinasingleplaneofanobject.

CT技术是使用X射线从不同方向对目标的某个平面进行扫描。

［5］Losslesscompressionmeansthatafileiscompactedwithoutlosinganyinformation,sothatthereconstructedfileisidenticaltotheoriginal．

无损压缩是不损失任何信息地将文件进行压缩,重建得到的文件和原文件是完全一样的。

［6］JPEG,anextremelycommonimagecompressionscheme,usesthediscretecosinetransform(DCT)toconcentratemostoftheinformationaboutan17×17sub-blockofanimageintoafewsignificantcoefficients.

JPEG是一种极其常用的图像压缩方法,该方法使用离散余弦变换将图像中17×17小块的大部分信息集中到少数几个重要系数上。

Exercises

1.PleasetranslatethefollowingphrasesintoEnglish.

(1)灰度图像(2)直方图均衡

(3)图像对比度 (4)结果矩阵

(5)边缘检测 (6)行程编码

(7)无损压缩 (8)低通滤波器

(9)哈夫曼编码 (10)数字图像压缩

2.PleasetranslatethefollowingphrasesintoChinese.

(1)edgefilter

(2)magnitudespectrum

(3)objectrecognition

(4)phasespectrum

(5)CT(ComputedTomography)

(6)MRI(MagneticResonanceImaging)

(7)DCT(DiscreteCosineTransform)

(8)highpassfilters

(9)lossycompression

3.TranslatethefollowingsentencesintoEnglish.

(1)人眼具有这样的特性:图像出现在视网膜上会保留几毫秒,然后消失。如果一个图像序列以50幅图像每秒逐行扫描显示,人们就不会觉得看到的是一幅离散的图像。所有视频系统都是利用这一原理产生运动画面的。

(2)所有压缩系统均需要两种算法:一个在信源压缩数据,一个在信宿对数据进行解压。在文献当中,这两个算法分别被称做编码算法和解码算法。

(3)JPEG(联合图像专家组)用于压缩连续色调的静止图像(例如照片)。它是ITU、ISO和IEC共同支持的图像专家开发出来的。

(4)宽高比是图像的宽度和高度之比。常规电视的宽高比是4∶3。高清晰度电视使用16∶9的宽高比。摄像机常用1.85∶1或者2.35∶1的宽高比。

(5)MPEG代表运动图像专家组,它是对数字压缩格式视听信息(如电影、录像、音乐)编码的一组标准的统称。与其他视频、音频的编码格式相比,MPEG的主要优势是相同质量下的文件要小得多。这是因为MPEG使用了非常复杂的压缩技术。

4.Answerthequestions.

(1)Whatisthefunctionofsubtractionofimages?

(2)Whatisthefunctionofwarping?

(3)Whatisdilation?

(4)Comparelosslesscompressiontolosscompression.

PassageBCompression/DecompressionTechniques

Numerousmethodshavebeendevelopedforthecompressionofdigitalimagedata.Oneoftheprincipaldriversforthisdevelopmentisthetelevisionindustrywherequalityimagedatamustbetransferredtoreceiversusingrelativelysimpleequipment.Thedevelopmentofhighdefinitiontelevisionisfurtherfocusingtheattentionofindustryanduniversityscientiststowardproblemsofdatareductionanddigitaltransmission.

Theprincipalevaluationcriteriafortheanalysisofcompressedversusuncompressedimageryiswhetherapersoncantellthedifferencebetweentheimages.AmoreimplementalmeasureistheRootMeanSquare(RMS)errorbetweentheoriginalimageandtheimagethathasbeencompressed.Compressionratesmaybegeneratedbydeterminingthesizeofthecompressedimageintermsofnumberofbitsperimagepixelfortheoriginalimage.[1]

Hereweonlyconsiderscompressionofsinglehighresolutionmulti-spectralimages.Highercompressionrateswillbeachievedinamotionsequencewhereframetoframevariationsmaybequantifiedandonlythechangesfromareferenceimageneedbecoded.

Therearetwogeneraltypesofcompression:(1)loss-less,and(2) loss.Loss-lesscompressionmeansthatyoucanachieveacertaincompressionfactorandbeabletoexactlyreproducetheoriginalimage.Losscompressionontheotherhandallowssomeloss,buthasthepotentialformuchhighercompressionrates.Nomatterwhattechniquethatyouuse,theexactrateisverydependentonthecomplexityoftheimagethatyouareanalyzing.Forexample,thenormalbestthatcanbeachievedwithloss-lessencodinginarateof2bitsperpixel.

Infact,forsomeLand-satsceneswithurbanareasandmanysmallfarms,thefactorof2bitsperpixelmaynotbeabletoachieved.ThesametechniqueappliedtoaLand-satimageofthemid-westwherelargefieldsoccurandfewshadowsexistinimagesmightproduceamuchbettercompression.

Oneloss-lesstechniqueisknownasrunlengthencoding.Thecompressionalgorithmprocesseseachlineofinputimagerylookingforregionsinwhichdatavaluesarethesame.Iftenpixelsintheoriginalimagehaveavalueof10,thenthesamedatamayberepresentedasadatavalue,10,andamultipliersayinghowmanytimesthevalueisrepeatedbeforeachangedvalue.Huffmanencodingfollowsasimilarprocess.Theseloss-lesstechniquesaregenerallycalledentropycodingtechniques,andhaveapplicationindocumentimaging,desktoppublishing,andGIS.Itshouldbenotedthatentropycodingdoesnotworkexceptionallywellintherepresentationofremotesensingimages.

Inremotesensingimageryitiswellknownthattheremaybesignificantcorrelationbetweendifferentbandsofmulti-spectraldata.Inimageprocessing,aprocedurecalledprincipalcomponentshasbeendesignedtoidentifycorrelationbetweenimagebandsandtocreateanewsetoftransformedbandsthatrepresentanewcolorspaceinwhichthenewimagebandsareuncorrelated.[2]Theprocedurealsoprovidesameasureofthepercentoftheoriginalvariationpresentintheoriginalimageasfoundineachofthenewtransformedbands.ForLand-satTMdata,threetofourofthetransformedimagesrepresent98percentofthevarianceintheoriginalimages;therefore,acompressionfactorof2wouldbeachievedwithlittleloss.

Anothertypeoftransformcodingdoesnotinvolvearotationofthecolorspace,butinsteadrepresentsimagesintermsofspatialfrequencyofcertainbasefunctions.Fouriertransformsmapanimageintoaspatialfrequencyimagebaseonsinandcosinefunctions.AfastcomputerimplementationoftheDiscreteFourierTransform(DFT)isknownasaFastFourierTransform(FFT).DiscreteCosineTransforms(DCT’s)mapthesameimagetoaspatialfrequencyimagebasedonlyonthecosinefunction.

Eachpixelmayberepresentedbyaseriesoftrigonometricfunctionsandcoefficientsderivedfromtheimages.Ifalltermsofthetransform’strigonometricfunctionsareused,compressionisminimal.Asmoretermsaredeleted,compressiongoesup,buttheresultingcompressedimagedevelopscertainartifactsoftheprocedure.

VectorQuantification(VQ)isatypeofencodingthatdefinesavectorrepresentationofnon-overlappingareablockswithinanimage.Avectorconsistsofvaluesrepresentingthedatavaluesforeachpixelwithintheregion.Usingthesevectors,clustersofvectorsarederivedusingaderivedspectraldistancemeasure.[3]Acodebookconsistingoftheclusteredvectorsisstored,representingthecharacteristicsoftheimage.Thisprocessisnumericallyintensiveandmaybeiterative.ThedecoderforVQtakesanimagevectorandcomparesittostoredvectorsinthecodebook.

Aselectionismadebasedonminimizationofadistortionfunctionbetweenthenewvectorandthecodebook.TheVQtechniquecangenerallyachievecompressionratiosofbetween20tooneand35toonewithlittleobservabledistortion.TheVQtechniqueisanexampleofasymmetriccompressioninthatconsiderablymoretimeisspentderivingthecodebookthanindecompressingviaacodebooklookup.Sincedifferentimagesmayhavedifferentcharacteristics,arobustcodebookisnecessarytosuccessfullycodeanddecodeLand-satandothersatelliteimages.AVQtechniqueusingbetweenchannelcorrelationaswellasspatialcorrelationachieveshighercompressionrateswithlosslessthanindependentbandVQ.

Thelasttypeofcompressionconsideredisfractalcompression.BasedonMandelbrotsetswhichtakeadvantageofaselfsimilar,scalingindependent,statisticalfeatureofnature,fractalcompressionanddecompressioninvolvesaclusteringapproachtofindingregionswhichexhibitthesamecharacteristicsasasampleregionwithoutregardtorotationandscale.Regionswithinanimagearerelatedtonumerousotherregionswithinthesameimage,withthisduplicationofinformationbeingthebasisofthecompressionpotential.

Fractalcompressioncanachievecompressionratiosofupto80toonewithonlymoderatelossofinformation.Thefractaltechnique,liketheVQtechniqueisalsoasymmetric.Hardwareimplementationofthedecompressionoffractalimageshasachievedreal-timerates.

NOTES

[1] Compressionratesmaybegeneratedbydeterminingthesizeofthecompressedimageintermsofnumberofbitsperimagepixelfortheoriginalimage.

压缩比可以通过原始图像每像素所需压缩的比特数来确定。

·intermsof…意为“根据…”。

[2] Inimageprocessing,aprocedurecalledprincipalcomponentshasbeendesignedtoidentifycorrelationbetweenimagebandsandtocreateanewsetoftransformedbandsthatrepresentanewcolorspaceinwhichthenewimagebandsareuncorrelated.

在图像处理过程中，一个称为主分量的过程被设计用以确认图像频谱之间的相关性，并产生一组新的变换频谱，其中新图像频谱互不相关地表示一个新的彩色空间。

·that引导定语从句修饰bands。inwhich引导定语从句修饰space。

[3] Usingthesevectors,clustersofvectorsarederivedusingaderivedspectraldistancemeasure.

使用这些矢量，用一个导出的频谱距离可推导出矢量簇。

·第二个using相当于byusing。

EXERCISES

1. True/False.

(1) Loss-lessandlossaretwogeneraltypesofcompression.()

(2) Nomatterwhattechniquethatyouuse,theexactcompressionrateisnotverydependentonthecomplexityoftheimagethatyouareanalyzing.()

(3) ThedecoderforVQtakesanimagevectorandcomparesittostoredvectorsinthecodebook.()

2.Fillintheblanks.

(1) Loss-lesscompressionmeansthatyoucanachieveacertain

andbeabletoexactlyreproducetheoriginalimage.

(2)AfastcomputerimplementationoftheDiscreteFourierTransformisknownasa

.

(3)VectorQuantification(VQ)isatypeofencodingthatdefinesa

ofnon-overlappingareablockswithinanimage.

(4) Thefractaltechnique,liketheVQtechniqueisalso

.

(5)

compressioncanachievecompressionradiosofupto80toonewithonlymoderatelossofinformation.

3. Chosethebestanswer.

(1) Whatkindofcompressionalgorithmprocesseseachlineinputimagerylookingforregionsinwhichdatavaluesarethesame?

a.runlengthencoding b.VectorQuantification

c.fractalcompression

(2) Whichtypeofcompressiontakeadvantageofaselfsimilar,scalingindependentandstatisticalfeatureofnature?

a.runlengthencoding b.VectorQuantification

c.fractalcompression

PassageCTelevision

Atelevisionpictureisbuiltupgraduallybymovingaspotoflightacrossanddownascreeninarasterpattern.

Thevideosignalcausesthebrightnessofthespottovaryinproportiontotheintensityoflightintheoriginalimage.Themovementofthespotacrossthescreeniscontrolledbythelinescansignal.Eachtimethespotreachestherightsideofthescreen,itisblankedandmovedrapidlybacktotheleftsidereadytostartthenextline.Thisrapidmovementbacktoastartingpositionisknownasfly-back.Eachcompleteimageorframerequiresaminimumof500linestogiveapictureofacceptablequality.ThepresentEuropeanTVsystemuses625linesperframe.

Themovementofthespotdownthescreeniscontrolledbythefieldscansignal.Whenthespotreachesthebottomofthescreen,itisblankedandmovedrapidlybacktothetopofthescreen.Theframemustbescannedatleastfortytimespersecondtopreventthescreenfromflickering.ThepresentEuropeanTVsystemhasaframescanrateof50 Hz.

ThevideosignalcontainslineandfieldsyncpulsestomakesurethattheTVreceiverstartsanewlineandanewframeatthesametimeastheTVcamera.

Toallowthevideosignaltobetransmittedusingasmallerrangeoffrequencies,eachframeistransmittedintwoseparatehalves,knownasfields.Thefirsttimethespottravelsdownthescreenitdisplaysthefirstfield,whichconsistsoftheodd-numberedframelines.Thesecondtimethespottravelsdownthescreenitdisplaysthesec