《信息工程專業(yè)英語》課件第6章

Unit6DigitalImageProcessing6.1Text6.2ReadingMaterials

6.1Text

FundamentalStepsinDigitalImageProcessing

Itishelpfultodividethematerialcoveredinthefollowingchaptersintothe

twobroadcategories:methodswhoseinputandoutputareimages,andmethodswhoseinputsmaybeimages,butwhoseoutputsareattributesextractedfromthoseimages.ThisorganizationissummarizedinFig6.1.Thediagramdoesnotimplythateveryprocessisappliedtoanimage.Rather,theintentionistoconveyanideaofallthemethodologiesthatcanbeappliedtoimagesfordifferentpurposesandpossiblywithdifferentobjectives.

ImageacquisitionisthefirstprocessshowninFig6.1.Notethatacquisitioncouldbeassimpleasbeinggivenanimagethatisalreadyindigitalform.Generally,theimageacquisitionstageinvolvespreprocessing,suchasscaling.

Fig6.1FundamentalstepsinDigitalImageProcessing

Imageenhancementisamongthesimplestandmostappealingareasofdigitalimageprocessing.Basically,theideabehindenhancementtechniquesistobringoutdetailthatisobscured,orsimplytohighlightcertainfeaturesofinterestinanimage.Afamiliarexampleofenhancementiswhenweincreasethecontrastofanimagebecause“itlooksbetter”.Itisimportanttokeepinmindthatenhancementisaverysubjectiveareaofimageprocessing.

Imagerestorationisanareathatalsodealswithimprovingtheappearanceofanimage.AnexampleofimagerestorationisshowninFig6.2.However,unlikeenhancement,whichissubjective,imagerestorationisobjective,inthesensethatrestorationtechniquestendtobebasedonmathematicalorprobabilisticmodelsofimagedegradation.Enhancement,ontheotherhand,isbasedonhumansubjectivepreferencesregardingwhatconstitutesa“good”enhancementresult.

Fig6.2Exampleofimagerestoration

ColorimageprocessingisanareathathasbeengaininginimportancebecauseofthesignificantincreaseintheuseofdigitalimagesovertheInternet.Waveletsarethefoundationforrepresentingimagesinvariousdegreesofresolution.

Compression,asthenameimplies,dealswithtechniquesforreducingthestoragerequiredtosaveanimage,orthebandwidthrequiredtotransmitit.Althoughstoragetechnologyhasimprovedsignificantlyoverthepastdecade,thesamecannotbesaidfortransmissioncapacity.ThisistrueparticularlyinusesoftheInternet,whicharecharacterizedbysignificantpictorialcontent.Imagecompressionisfamiliar(perhapsinadvertently)tomostusersofcomputersintheformofimagefileextensions,suchasthejpgfileextensionusedintheJPEG(JointPhotographicExpertsGroup)imagecompressionstandard.

Morphologicalprocessingdealswithtoolsforextractingimagecomponentsthatareusefulintherepresentationanddescriptionofshape.Thematerialin

thischapterbeginsatransitionfromprocessesthatoutputimagestoprocesses

thatoutputimageattributes.

Segmentationprocedurespartitionanimageintoitsconstituentpartsorobjects.AnexampleofimagesegmentationisshowninFig6.3.Ingeneral,autonomoussegmentationisoneofthemostdifficulttasksindigitalimageprocessing.Aruggedsegmentationprocedurebringstheprocessalongwaytowardsuccessfulsolutionofimagingproblemsthatrequireobjectstobeidentifiedindividually.Ontheotherhand,weakorerraticsegmentationalgorithmsalmostalwaysguaranteeeventualfailure.Ingeneral,themoreaccurate

thesegmentation,themorelikelyrecognitionistosucceed.

Fig6.3Exampleofimagesegmentation

Representationanddescriptionalmostalwaysfollowtheoutputofasegmentationstage,whichusuallyisrawpixeldata,constitutingeithertheboundaryofaregion(i.e.,thesetofpixelsseparatingoneimageregionfromanother)orallthepointsintheregionitself.Ineithercase,convertingthedatatoaformsuitableforcomputerprocessingisnecessary.Thefirstdecisionthatmustbemadeiswhetherthedatashouldberepresentedasaboundaryorasacompleteregion.

Boundaryrepresentationisappropriatewhenthefocusisonexternalshapecharacteristics,suchascornersandinflections.Regionalrepresentationisappropriatewhenthefocusisoninternalproperties,suchastextureorskeletalshape.Insomeapplications,theserepresentationscomplementeachother.Choosingarepresentationisonlypartofthesolutionfortransformingrawdataintoaformsuitableforsubsequentcomputerprocessing.Amethodmustalsobespecifiedfordescribingthedatasothatfeaturesofinterestarehighlighted.Description,alsocalledfeatureselection,dealswithextractingattributesthatresultinsomequantitativeinformationofinterestorarebasicfordifferentiatingoneclassofobjectsfromanother.

Recognitionistheprocessthatassignsalabel(e.g.,“vehicle”)toanobjectbasedonitsdescriptors.Weconcludeourcoverageofdigitalimageprocessingwiththedevelopmentofmethodsforrecognitionofindividualobjects.

Knowledgeaboutaproblemdomainiscodedintoanimageprocessingsystemintheformofaknowledgedatabase.Thisknowledgemaybeassimpleasdetailingregionsofanimagewheretheinformationofinterestisknowntobelocated,thuslimitingthesearchthathastobeconductedinseekingthatinformation.Theknowledgebasealsocanbequitecomplex,suchasaninterrelatedlistofallmajorpossibledefectsinamaterialsinspectionproblemoranimagedatabasecontaininghigh-resolutionsatelliteimagesofaregioninconnectionwithchange-detectionapplications.Inadditiontoguidingtheoperationofeachprocessingmodule,theknowledgebasealsocontrolstheinteractionbetweenmodules.ThisdistinctionismadeinFig6.1bytheuseofdouble-headedarrowsbetweentheprocessingmodulesandtheknowledgebase,asopposedtosingle-headedarrowslinkingtheprocessingmodules.

Althoughwedonotdiscussimagedisplayexplicitlyatthispoint,itisimportanttokeepinmindthatviewingtheresultsofimageprocessingcantakeplaceattheoutputofanystageinFig6.1.WealsonotethatnotallimageprocessingapplicationsrequirethecomplexityofinteractionsimpliedbyFig6.1.Infact,notevenallthosemodulesareneededinsomecases.Forexample,imageenhancementforhumanvisualinterpretationseldomrequiresuseofanyoftheotherstagesinFig6.1.Ingeneral,however,asthecomplexityofanimageprocessingtaskincreases,sodoesthenumberofprocessesrequiredtosolvetheproblem.

Technicalwordsandphrases

broad adj.顯著的;寬的，遼闊的

attribute n.屬性；特質(zhì)

scaling

n.縮放比例

obscured

v.使含混；變得模糊

restoration n.恢復(fù)；復(fù)原；歸還

degradation n.退化；降格，降級

resolution n.分辨率

pictorial adj.圖像的；繪畫的

inadvertently adv.非故意地；不注意地

morphological adj.形態(tài)學(xué)的

rugged adj.崎嶇的；堅固的；高低不平的

erratic

adj.不穩(wěn)定的

raw adj.生的；未加工的

inflection n.曲線，彎曲，變形

texture n.紋理；質(zhì)地

skeletal

adj.骨骼的，像骨骼的

complement vt.補足，補助

quantitative adj.定量的；量的，數(shù)量的

differentiate vi.區(qū)分，區(qū)別

inspection

n.視察，檢查

defect

n.缺點，缺陷；不足之處

interaction

n.相互作用；互動

distinction

n.特性；區(qū)別；差別；榮譽、勛章

explicit

adj.明確的；清楚的；直率的；詳述的

interpretation n.解釋；翻譯；演出

beextractedfrom

從……中提取

contrastofanimage

圖像對比度

mathematicalmodels 數(shù)學(xué)模型

probabilisticmodels 概率模型

autonomoussegmentation 自動分割

partitioninto

劃分

high-resolutionsatelliteimages 高分辨度衛(wèi)星圖像

double-headedarrows

雙箭頭

asopposedto

與……相反

JPEG

(JointPhotographicExpertsGroup)聯(lián)合圖片專家組

6.1.1Exercises

1.PutthePhrasesintoEnglish

(1)數(shù)碼圖像處理； (2)彩色圖像；

(3)存儲技術(shù)； (4)傳輸能力；

(5)文件擴展名； (6)分割算法；

(7)原始數(shù)據(jù)； (8)后續(xù)處理；

(9)特征選擇； (10)個別目標(biāo)。

2.PutthePhrasesintoChinese

(1)digitalimageprocessing;

(2)imageacquisition;

(3)imageenhancement;

(4)highlightcertainfeatures;

(5)knowledgedatabase;

(6)descriptionofshape;

(7)assignsalabelto;

(8)aproblemdomain.

3.Translation

(1)Basically,theideabehindenhancementtechniquesistobringoutdetailthatisobscured,orsimplytohighlightcertainfeaturesofinterestinanimage.

(2)Inthesensethatrestorationtechniquestendtobebasedonmathematicalorprobabilisticmodelsofimagedegradation.

(3)Aruggedsegmentationprocedurebringstheprocessalongwaytowardsuccessfulsolutionofimagingproblemsthatrequireobjectstobeidentifiedindividually.

(4)Thefirstdecisionthatmustbemadeiswhetherthedatashouldberepresentedasaboundaryorasacompleteregion.

(5)Theknowledgebasealsocanbequitecomplex,suchasaninterrelatedlistofallmajorpossibledefectsinamaterialsinspectionproblemoranimagedatabasecontaininghigh-resolutionsatelliteimagesofaregioninconnectionwithchange-detectionapplications.

(6)Ingeneral,however,asthecomplexityofanimageprocessingtaskincreases,sodoesthenumberofprocessesrequiredtosolvetheproblem.

6.1.2參考譯文

將圖像處理劃分為兩個主要類別：一類是其輸入和輸出都是圖像，另一類，輸入可能是圖像，但是輸出是從圖像中提取的特征屬性。這一結(jié)構(gòu)在圖6.1中做了概括。該圖并不意味著每一個處理步驟都應(yīng)用于一張圖片，其意圖是要說明針對不同目標(biāo)的圖像處理，這些方法都可應(yīng)用其中。

圖像獲取是圖6.1的第一步處理。注意圖像獲取與給出一幅數(shù)字形式的圖像一樣簡單。通常，圖像獲取包括預(yù)處理，如圖像的縮放。

圖像增強是數(shù)碼圖像處理最簡潔和最有吸引力的環(huán)節(jié)?；旧希鰪娂夹g(shù)的基本思路就是顯現(xiàn)那些被模糊了的細節(jié)，或通過簡化來突出圖像中某個感興趣的特征。一個關(guān)于圖像增強熟悉的例子：當(dāng)我們增加圖像的對比度，它就會更好看。重要的是要記住，圖像增強是圖像處理中非常主觀的領(lǐng)域。

圖像復(fù)原是改進圖像外觀的環(huán)節(jié)。圖像復(fù)原的例子如圖6.2所示。然而，不像圖像增強是主觀的，圖像復(fù)原是客觀的。從某種意義上說，復(fù)原技術(shù)是基于圖像退化的數(shù)學(xué)模型或概率模型。而圖像增強從另一方面來看，是以人的主觀偏愛為基礎(chǔ)來得到“好”的增強效果。

彩色圖像處理已經(jīng)變得越來越重要，是因為互聯(lián)網(wǎng)上數(shù)碼圖像的應(yīng)用有著顯著的增長。微波就是在各種分辨率下描述圖像的基礎(chǔ)。

壓縮，顧名思義，即減少圖像的存儲量，或者降低傳輸圖像的帶寬。雖然存儲技術(shù)在過去的10年內(nèi)有了顯著提升，但傳輸容量并非如此。以大量圖片內(nèi)容為特征的互聯(lián)網(wǎng)更是如此。圖像壓縮技術(shù)所對應(yīng)的圖像文件擴展名對大多數(shù)計算機用戶來說是很熟悉的(有的也許并不注意)，如在JPEG(聯(lián)合圖片專家組)圖像壓縮標(biāo)準(zhǔn)中使用JFG文件擴展名。

形態(tài)學(xué)處理會涉及到提取圖像元素，這些圖像元素在表現(xiàn)和描述形狀方面非常有用。這一章的材料將從輸出圖像處理到輸出圖像特征處理的轉(zhuǎn)換開始。

分割過程是將一幅圖像劃分為組成圖像的各部分或目標(biāo)物，圖像分割的例子如圖6.3所示。通常，自動分割是數(shù)字圖像處理中最為困難的任務(wù)之一。復(fù)雜的分割過程導(dǎo)致成功解決分割問題變得很難，這要求物體被分別識別出來，這種分割成像的問題需要大量的處理工作。另一方面，不健全且缺乏穩(wěn)定的分割算法幾乎總是會導(dǎo)致最終的失敗。通常，分割越準(zhǔn)確，識別越成功。

表示與描述幾乎總是跟隨在分割步驟的輸出后邊，通常這樣的輸出是未加工的像素數(shù)據(jù)，其構(gòu)成不是區(qū)域的邊緣(也就是分隔一個圖像區(qū)域和另一個區(qū)域的像素集)就是其區(qū)域本身所有點的集合。無論哪種情況，把數(shù)據(jù)轉(zhuǎn)換成適合計算機處理的形式都是必要的。首先，必須確定數(shù)據(jù)是被表現(xiàn)為邊界還是整個區(qū)城。當(dāng)注意的焦點是外部形狀特性時，邊界表示是合適的，如拐角和曲線。當(dāng)注意的焦點是內(nèi)部特性時，則區(qū)域表示是合適的，如紋理或骨骼形狀。某些應(yīng)用中，這些表示相輔相成的。選擇某種表現(xiàn)方式僅是把原始數(shù)據(jù)轉(zhuǎn)換為適合計算機后續(xù)處理形式過程中的一部分。為了通過描述數(shù)據(jù)使感興趣的特征表現(xiàn)得更明顯，還必須確定一種方法。描述也叫特征選擇，涉及提取特征，該特征是某些感興趣的定量信息或是區(qū)分一組目標(biāo)與其他目標(biāo)的基礎(chǔ)。

識別是在目標(biāo)描述基礎(chǔ)上給目標(biāo)賦予標(biāo)簽(例如“車輛”)的過程。我們用個別目標(biāo)識別的先進方法來推導(dǎo)數(shù)字圖像處理的覆蓋范圍。

問題域是將問題以知識庫的形式編碼并裝入某個圖像處理系統(tǒng)。這一過程就如詳述感興趣的信息位于某個圖像區(qū)域那樣簡單，這樣限制性的搜索就被引導(dǎo)到要尋找的信息處。知識庫也可能相當(dāng)復(fù)雜，如材料檢測問題或者圖像數(shù)據(jù)庫中所有主要缺陷的相關(guān)列表，數(shù)據(jù)庫包含變化檢測應(yīng)用相關(guān)區(qū)域的高分辨衛(wèi)星圖像。除了引導(dǎo)每個處理模塊的操作，知識庫還要控制模塊之間的交互。這種特性由圖6.1中的處理模塊和知識庫之間的雙向箭頭表示，這與單向箭頭連接處理模塊截然相反。

此時此刻我們雖然沒有明確地討論圖像顯示，但要記住觀察圖6.1中任何階段輸出處的圖像處理結(jié)果是很重要的。我們還注意到，不是所有的圖像處理都需要圖6.1所給出的復(fù)雜交互。事實上，在某些情況下并不是所有模塊都需要。例如，為了人的視覺解釋，圖像增強很少需要使用圖6.1中的其他任何步驟。然而，通常隨著圖像處理任務(wù)復(fù)雜度的增加，則需要做更多處理才能使問題得到解決。

6.2ReadingMaterials

6.2.1PyramidMethodsinImageProcessing

Digitalimageprocessingisbeingusedinmanydomainstoday.Inimageenhancement,forexample,avarietyofmethodsnowexistforremovingimagedegradationsandemphasizingimportantimageinformation,andincomputergraphics,digitalimagescanbegenerated,modified,andcombinedforawidevarietyofvisualeffects.Indatacompression,imagesmaybeefficientlystoredandtransmittediftranslatedintoacompactdigitalcode.Inmachinevision,automaticinspectionsystemsandrobotscanmakesimpledecisionsbasedonthedigitizedinputfromatelevisioncamera.

Thetaskofdetectingatargetpatternthatmayappearatanyscalecanbeapproachedinseveralways.Twoofthese,whichinvolveonlysimpleconvolutions,areillustratedinFig6.4.Severalcopiesofthepatterncanbeconstructedatincreasingscales,andtheneachisconvolvedwiththeimage.Alternatively,apatternoffixedsizecanbeconvolvedwithseveralcopiesoftheimagerepresentedatcorrespondinglyreducedresolutions.Thetwoapproachesyieldequivalentresults,providedcriticalinformationinthetargetpatternisadequatelyrepresented.However,thesecondapproachismuchmoreefficient:

agivenconvolutionwiththetargetpatternexpandedinscalebyafactorswillrequires4morearithmeticoperationsthanthecorrespondingconvolutionwiththeimagereducedinscalebyafactorofs.Thiscanbesubstantialforscalefactorsintherange2to32,acommonlyusedrangeinimageanalysis.

Fig6.4Twomethodsofsearchingforatargetpatternovermanyscales

Theimagepyramidisadatastructuredesignedtosupportefficientscaledconvolutionthroughreducedimagerepresentation.Itconsistsofasequenceofcopiesofanoriginalimageinwhichbothsampledensityandresolutionaredecreasedinregularsteps.AnexampleisshowninFig6.5a.Thesereducedresolutionlevelsofthepyramidarethemselvesobtainedthroughahighlyefficientiterativealgorithm.Thebottom,orzerolevelofthepyramid,G0,isequaltotheoriginalimage.

Thisislowpassfilteredandsubsampledbyafactoroftwotoobtainthenextpyramidlevel,Gl.GlisthenfilteredinthesamewayandsubsampledtoobtainG2.Furtherrepetitionsofthefilter/subsamplestepsgeneratetheremainingpyramidlevels.Tobeprecise,thelevelsofthepyramidareobtainediterativelyasfollows.For0<l<N:

(6-2-1)

However,itisconvenienttorefertothisprocessasastandardREDUCEoperation,andsimplywrite:Gl=REDUCE[Gl-1].

Wecalltheweightingfunctionω(m,n)the“generatingkernel”.Forreasonsofcomputationalefficiencythisshouldbesmallandseparable.Afive-tapfilterwasusedtogeneratethepyramidinFig6.5.

Fig6.5TheGaussianPyramidexpandedtothesizeoftheoriginalimage

PyramidconstructionisequivalenttoconvolvingtheoriginalimagewithasetofGaussian-likeweightingfunctions.These“equivalentweightingfunctions”forthreesuccessivepyramidlevelsareshowninFig6.6.Notethatthefunctionsdoubleinwidthwitheachlevel.Theconvolutionactsasalowpassfilterwiththebandlimitreducedcorrespondinglybyoneoctavewitheachlevel.BecauseofthisresemblancetotheGaussiandensityfunctionwerefertothepyramidoflowpassimagesasthe“Gaussianpyramid.”

Fig6.6Equivalentweightingfunctions

Bandpass,ratherthanlowpass,imagesarerequiredformanypurposes.ThesemaybeobtainedbysubtractingeachGaussian(lowpass)pyramidlevelfromthenextlowerlevelinthepyramid.Becausetheselevelsdifferintheirsampledensityitisnecessarytointerpolatenewsamplevaluesbetweenthoseinagivenlevelbeforethatlevelissubtractedfromthenext-lowerlevel.InterpolationcanbeachievedbyreversingtheREDUCEprocess.WecallthisanEXPANDoperation.LetGl,kbetheimageobtainedbyexpandingGlktimes.ThenGl,k=EXPAND[Gl,k-1]or,tobeprecise,Gl,0=Gl,andfork?>

0,

(6-2-2)

Hereonlytermsforwhich(2i+m)/2and(2j+n)/2areintegerscontributetothesum.Theexpandoperationdoublesthesizeoftheimagewitheachiteration,sothatGl,1,isthesizeofGl,1,andGl,1isthesamesizeasthatoftheoriginalimage.ExamplesofexpandedGaussianpyramidlevelsareshowninFig6.7.

Fig6.7LevelsoftheGaussianpyramidexpandedtothesizeoftheoriginalimage

Thelevelsofthebandpasspyramid,L0,L1,...,LN,maynowbespecifiedintermsofthelowpasspyramidlevelsasfollows:

(6-2-3)

ThefirstfourlevelsareshowninFig6.8.

Fig6.8Thelevelsofthebandpasspyramid,L0,L1,L2,L3

JustasthevalueofeachnodeintheGaussianpyramidcouldhavebeenobtaineddirectlybyconvolvingaGaussianlikeequivalentweightingfunctionwiththeoriginalimage,eachvalueofthisbandpasspyramidcouldbeobtainedbyconvolvingadifferenceoftwoGaussianswiththeoriginalimage.ThesefunctionscloselyresembletheLaplacianoperatorscommonlyusedinimageprocessing.Forthisreasonwerefertothebandpasspyramidasa“Laplacianpyramid.”

AnimportantpropertyoftheLaplacianpyramidisthatitisacompleteimagerepresentation:thestepsusedtoconstructthepyramidmaybereversedtorecovertheoriginalimageexactly.Thetoppyramidlevel,LN,isfirstexpandedandaddedtoLN-1toformGN-1thenthisarrayisexpandedandaddedtoLN-2torecoverGN-2,andsoon.Alternatively,wemaywrite:

(6-2-4)

Thepyramidhasbeenintroducedhereasadatastructureforsupportingscaledimageanalysis.Thesamestructureiswellsuitedforavarietyofotherimageprocessingtasks.Itcanbeshownthatthepyramid-buildingproceduresdescribedherehavesignificantadvantagesoverotherapproachestoscaledanalysisintermsofbothcomputationcostandcomplexity.ThepyramidslevelsareobtainedwithfewerstepsthroughrepeatedREDUCEandEXPANDoperationsthanispossiblewiththestandardFFT.Furthermore,directconvolutionwithlargeequivalentweightingfunctionsrequires20to30bitarithmetictomaintainthesameaccuracyasthecascadeofconvolutionswiththesmallgeneratingkernelusingjust8-bitarithmetic.

6.2.2ImageCompressionandCoding

Wepresentanalgorithmforimagecompressionbasedonanimageinpaintingmethod.Firsttheimageregionsthatcanbeaccuratelyrecoveredarelocated.Then,toreducethedata,informationofsuchregionsisremoved.Theremainingdatabesidesessentialdetailsforrecoveringtheremovedregionsareencodedtoproduceoutputdata.Atthedecoder,aninpaintingmethodisappliedtoretrieveremovedregionsusinginformationextractedattheencoder.

Theimageinpaintingtechniqueutilizespartialdifferentialequations(PDEs)forrecoveringinformation.Itisdesignedtoachievehighperformanceintermsofimagecompressioncriteria.Thisalgorithmwasexaminedforvariousimages.Ahighcompressionratioof1:40wasachievedatanacceptablequality.ExperimentalresultsshowedattainablevisiblequalityimprovementatahighcompressionratiocomparedwithJPEG.

Compressionisacceptablefornaturalimagesasalargeamountofredundancyisincludedinsuchimages.Ahighcompressionratiocanbeachievedbyeliminatingtheseredundancies,butatthecostofsomeinformationloss.Uptonow,greatachievementshavebeenmadeinimagecompression.State-of-the-artmethodssuchasJPEG(PennebakerandMitchell,1992)andJPEG2000(TaubmanandMarcellin,2002)efficientlyexploitstatisticalredundanciesamongpixelsandachievehighcompressionratios.

Acom-monframeworkinalmostalllossyimagecompressionmethodsisimagetransformationfollowedbyquantizationandcoding.Incontrast,JPEGandJPEG2000usethediscretecosinetransform(DCT)andwavelettransform,respectively.Informationlosscausesdecreaseinthequalityofreconstructedimages,especiallyathighcompressionratios.Toimproveperceptualvisualquality,HontschandKaram(2000)andMaloetal.(2006)incorporatedthehumanvisionsystem(HVS)propertiesincompressionschemes.

Thedominanttypeofredundancywithinimagescomesfromtheirrepresentationprocedure.Eachdigitalimageiscomposedofdiscretepoints,calledpixels.Thevaluerelevanttoeachpixelistheresultofsamplingfromlightorcolorintensityintheoriginalimagedomain.Naturalimagesconsistofseparateareasindicatingtheobjectsurfacesorsceneries.Be-causethelightintensityandcolorinsuchareasareapproximatelyconstant,therelevantvaluesforpixelsarehighlycorrelated.Everypixelinsuchareasislikelytobeofthesameorveryclosevaluecomparedwiththeadjacentpixels.Inthiscase,imagessufferfromahighlevelofspatialcorrelation.

Themostsignificantinformationwithinanimageislocatedintheboundaryregionsoredges.Infact,theboundaryofaregionnotonlyspecifiesitsoverallshape,butalsoshowshowpixelvalueschangefromneighboringregionstotheinnerregionsofinterest.Asaresult,itispossibletoretrievetheinnerareasusingpixelslocatedontheboundaries.Therefore,boundariesoredgesarealltherequiredinformationfordisplayinganimage.Fig6.9

clarifiesthisconcept.Fig6.9(a)showsasyntheticimagecomposedofthreedifferentregions.

Fig6.9(b)showstheboundariesofregionsinFig6.9(a).Certainly,bydiffusingtheinformationofboundariesshowninFig6.9(b)intothecorrespondingregion,theimageinFig6.9(a)willbere-covered.InFig6.9(b),theboundaryregionsarepreservedwith9pixelswidth.Also,theinformationrelatedtootherregionsisremoved.Itisevidentthatonly2pixelsratherthan9areadequateforrepresentingboundaryregions.

(a)Animagewiththreeregions;

(b)Extractededgesofthesameimagewith9pixelswidthFig6.9

Thevariationinvaluesofpixelsorthogonaltotheedgesissignificant.Hence,areasintheneighbor-hoodofedgesmaybeconsideredasessentialimageinformation.Ontheotherhand,whilemovingalongtheedgedirection,nosignificantchangesinpixelvalueswillbeobserved.Movingfurthertotheinnerpointsoftheboundarieswillresultinaconsiderablecorrelationforpixelvalues.Edgesalsorepresentsomeothernecessaryinformationincludingshapes.Redundanciesrelatedtothecorrelationalongtheedgedirectionmayalsobeexploitedviaextractingshapeinformationfrompixelvaluesinboundaryregions.

Pixelvaluesattheendpointsofanedgewillbeusedforrecoveringtheentireedgeandboundaryregion.Ofcourse,itholdstrueifthelocationoftheedgepointsisknown.Inthispaper,theedgeendpointiscalledthe'sourcepoint'.Inordertorecoverboundaryregionsandpixelslocatedperpendiculartotheedgedirection,samplesofsourcepointsshouldbeprovided.Thesesamplesshouldcomefromdifferentareasateachsideoftheedge.Fig6.10indicatesedgelocationsandsourcepointsfortheimageshowninFig6.10(a),zoomedversionisshowninFig6.10(b).

(a)AsourcepointsandboundariesforFig6.9a(theblacklinesstandforedges)Fig6.10(b)ZoomedversionofFig6.9aFig6.10

ZoomedversionofaJPEG2000istoadapttothecontinuousdevelopmentofimagecompressionapplications,theemergenceofnewstillimagecompressionstandard.JPEG2000imagecodingsystemdescribestherealizationoftheprocess,ofwhichthebasicalgorithmusedisdescribedandkeytechnologies,introducedthenewstandardfeaturesandapplications,anditsperformanceisanalyzed.Withtherapidgrowthofmultimediaapplicationsandnetworkscontinuestodevelop,thetraditionalJPEGcompressiontechnologyhasbeenunabletomeetthepeople'sdigitalmultimediaimagedatarequirements,amorepowerfulandefficientsuperiorstillimagecompressionstandardhasbeenreferredtothedevelopmentagenda,thisisaJPEG2000.

JPEG(JointPhotographicExpertsGroup)istheInternationalOrganizationforStandardization(ISO)developedundertheleadershipofthecommitteestillimagecompressionstandard,thefirstsetofinternationalstillimagecompressionstandardISO10918-1(JPEG)isthatestablishedbytheCommittee.AstheexcellentqualityJPEG,sothatwithinafewyearshewasagreatsuccess,iswidelyusedinthefieldofInternetanddigitalcamerasonthesite80%haveadoptedtheJPEGimagecompressionstandard.

However,thecurrentJPEGstillimagecompressionstandard,withmid-rangeandhigh-bit-rateontheratedistortioncharacteristicsofagood,butinthecontextoflowbitrate,therewillbeobviousblockingeffects,itsqualityhasbecomeunacceptable.JPEGcannotbeprovidedinasinglebitstreamlossyandlosslesscompression,andcannotsupportmorethan64×64Koftheimagecompression.Atthesametime,despitethecurrentJPEGstandardhasarequirementtorestarttheinterval,butwhenbiterrorsencounteredwhentheimagequalitywillbeseriouslydamaged.

Tosolvetheseproblems,sinceMarch1997onwards,JPEGimagecompressionstandardscommitteestartedtodevelopanewgenerationofimagecompressionstandardtoaddresstheseproblems.InMarch2000theTokyoconferencetoidentifyanewgenerationofcolorstillimagecodingstandardJPEG2000imagecompressioncodingalgorithm.

JPEG2000advantagesofitsuniquemakeupfordeficienciesintheexistingJPEGstandard.Discretewavelettransformalgorithm,theimagecanbeconvertedintoaseriesofpixelscanbemoreeffectivememorymodulesub-band,therefore,JPEG2000imagecompressionformatthantheJPEGcanbebasedonthecurrentre-increasedby10%to30%,andthecompressedimageappearstomoredelicatesmooth.Inotherwords,onlineviewingofimagesusingJPEG2000compression,notonlytodownloadspeedsfasterthanusingJPEGformat,nearly30%,butthequalitywillbebetter.

ForthecurrentJPEGstandard,inthesamecompressedstreamcannotprovidelossyandlosslesscompression,whileinJPEG2000systems,byselectingtheparameters,canbelossyandlosslessimagecompression,imagequalitytomeetthedemandingmedicalimages,imagelibrary,etc.processingneeds.NowthenetworkisbasedonJPEGimagedownload“block”transfers,soitcanonlybedisplayedlinebyline,whiletheuseofJPEG2000imageformatsupportforprogressivetransmission(ProgressiveTransmission),whichallowstheimageresolutionorpixelsinaccordancewiththerequiredaccuracyofreconstruction,theuserneededforimagetransmissio