[22]。Flask可以由很多配置的方法，甚至在不同的環(huán)境中也可以進(jìn)行配置，十分方便；其環(huán)境部署也很簡單：只需在安裝python和IDE的情況下，運(yùn)行命令行即可。在所以只需要在python中導(dǎo)入相應(yīng)的包即可。為了方便，可以直接下載anaconda來代替python；低耦合特點(diǎn)：Flask可以兼容多種數(shù)據(jù)庫和模板，因此具有低耦合的特點(diǎn)。Anaconda介紹Anaconda是一個可以用于科學(xué)計(jì)算的python發(fā)行版，是一個集成的python工具（自帶安裝很多庫）提供了包管理功能和環(huán)境管理功能，解決了多版本python并存、切換的問題。系統(tǒng)實(shí)現(xiàn)功能書法集圖像去噪介紹的展示功能如圖6所示：圖6.介紹部分功能塊上傳原始圖片功能該功能實(shí)現(xiàn)了將原始的圖像上傳的功能，進(jìn)而可以進(jìn)行下一步的操作。原始圖片代碼見代碼1：代碼1原始圖片上傳divdivclass="uisegment">

<divclass="uilabelteal">上傳書法圖片，系統(tǒng)自動實(shí)現(xiàn)去噪功能</div>

<p></p>

<formclass="uiform"action='./2/upload'enctype="multipart/form-data"method='post'onsubmit="showLoading()"name="form1">

<divclass="field">

<inputtype='file'name='file'id='file'accept='image/*'style="display:none;"onChange="document.form1.path.value=this.value"/>

<inputtype="button"value="selectimage"onclick="document.form1.file.click()">

<divclass="uicontainer"id='content'>

</div>

</div>上傳圖片界面如圖7所示：圖7.圖像上傳功能塊調(diào)用算法調(diào)用算法的過程在上傳完圖片后進(jìn)行。通過和去噪算法模型連接達(dá)到調(diào)用算法的目的。調(diào)用算法代碼如代碼2所示：代碼2調(diào)用算法varfile=document.getElementById('file');

varrealPath=getObjectURL(file.files[0]);

//document.getElementById('content').innerHTML=realPath;

img.src=realPath;

img.style.width="100%";

//img.style.height="200px";

},false);

functiongetObjectURL(file){

varurl=null

if(window.createObjectURL!=undefined){//basic

url=window.createObjectURL(file)

}elseif(window.URL!=undefined){//mozilla(firefox)

url=window.URL.createObjectURL(file)

}elseif(window.webkitURL!=undefined){//webkitorchrome

url=window.webkitURL.createObjectURL(file)

}

returnurl;

}生成去噪圖片的展示功能將去噪后的圖像進(jìn)行展示，并且展示出去噪前和去噪后的區(qū)別。生成圖片的展示功能過程見代碼3：代碼3生成圖片展示<<divclass="uigrid">

<divclass="column"style="width:200px;">

原圖

<imgsrc="{{static_url(img_path_ori)}}"width="200px"height="200px">

</div>

<divclass="column"></div>

<divclass="column"style="width:200px;">

處理后

<imgsrc="{{static_url(img_path_out)}}"width="200px"height="200px">

</div>

</div>生成界面結(jié)果如圖8所示：圖7.圖像上傳功能塊總結(jié)與展望總結(jié)本文從書法圖像修復(fù)的實(shí)際意義出發(fā)，在了解并學(xué)習(xí)了深度學(xué)習(xí)算法模型的基礎(chǔ)上，基于flask框架設(shè)計(jì)了書法圖像去噪系統(tǒng)，并調(diào)用通過深度學(xué)習(xí)進(jìn)行去噪的書法圖像去噪模型。設(shè)計(jì)用戶界面模塊，讓人們了解到圖像去噪對書法圖像保護(hù)的意義，實(shí)現(xiàn)在線圖片上傳以及通過圖像去噪后的結(jié)果圖片展示。未來展望本系統(tǒng)只是單一的對書法圖像進(jìn)行去噪，功能較為單一。因?yàn)槭菆D像去噪的相關(guān)內(nèi)容，可以在此基礎(chǔ)上上傳多個算法模型的去噪結(jié)果進(jìn)行比較，讓人們在得到去噪圖片的基礎(chǔ)上對算法模型有更加升入的了解。參考文獻(xiàn)LiuH,ZhouN.Animprovedfilteringalgorithmbasedonmedianfilteringalgorithmandmediumfilteringalgorithm[C]//AdvancedComputationalIntelligence(ICACI),2012IEEEFifthInternationalConferenceon.IEEE,2012:574-578.KandemirC,KalyoncuC,Toygar?.Aweightedmeanfilterwithspatial-biaseliminationforimpulsenoiseremoval[J].DigitalSignalProcessing,2015,46:164-174.GuemriK,DriraF.Adaptativeshockfilterforimagecharactersenhancementanddenoising[C]//SoftComputingandPatternRecognition(SoCPaR),20146thInternationalConferenceof.IEEE,2014:279-283.NguyenHT,Linh-TrungN.TheLaplacianpyramidwithrationalscalingfactorsandapplicationonimagedenoising[C]//InformationSciencesSignalProcessingandtheirApplications(ISSPA),201010thInternationalConferenceon.IEEE,010:468-471.WangSZ,LeeHJ.Dual-binarizationandanisotropicdiffusionofChinesecharactersincalligraphydocuments[C]//DocumentAnalysisandRecognition,2001.Proceedings.SixthInternationalConferenceon.IEEE,2001:271-275.DabovK,FoiA,KatkovnikV,etal.Imagedenoisingbysparse3-Dtransform-domaincollaborativefiltering[J].IEEETransactionsonimageprocessing,2007,16(8):2080-2095.KrizhevskyA,SutskeverI,HintonGE.Imagenetclassificationwithdeepconvolutionalneuralnetworks[C]//Advancesinneuralinformationprocessingsystems.2012:1097-1105.GOODFELLOWIJ,POUGET-ABADIEJ,MIRZAM,etal.Generativeadversarialnets[C]//Internationalconferenceonneuralinformationprocessingsystems.Massachusetts:MIT,2014:2672-2680.ZhangY,TianY,KongY,etal.Residualdensenetworkforimagesuper-resolution[C]//TheIEEEConferenceonComputerVisionandPatternRecognition(CVPR).2018.RAJEEVR,SAMATHJA,KARTHIKEYANNK.Anintelligentrecurrentneuralnetworkwithlongshort-termmemory(LSTM)BASEDbatchnormalizationformedicalimagedenoising[J].JournalofMedicalSystems,2019,43(8):1-10.TRIPATHIPC,BAGS.CNN-DMRI:aconvolutionalneuralnetworkfordenoisingofmagneticresonanceimages[J].PatternRecognitionLetters,2020,135:57-63.RadfordA,MetzL,ChintalaS.UnsupervisedRepresentationLearningwithDeepConvolutionalGenerativeAdversarialNetworks[J].E-printarXiv:1511.06434,2015:1-15.MirzaM,OsinderoSConditionalgenerativeadversarialnets[J].arXivpreprintarXiv:1411.1784,2014.IsolaP,ZhuJY,ZhouT,etal.Image-To-ImageTranslationWithConditionalAdversarialNetworks[C]//ProceedingsoftheIEEEConferenceonComputerVisionandPatternRecognition.2017:1125-1134.RonnebergerO,FischerP,BroxT.U-net:Convolutionalnetworksforbiomedicalimagesegmentation[C]//InternationalConferenceonMedicalimagecomputingandcomputer-assistedintervention.Springer,Cham,2015:234-241.ZhuJY,ParkT,IsolaP,etal.UnpairedImage-to-ImageTranslationusingCycle-ConsistentAdversarialNetworks[C]//IEEEInternationalConferenceonComputerVision.2017.ArjovskyM,ChintalaS,BottouL.Wassersteingenerativeadversarialnetworks[C]//InternationalConferenceonMachineLearning.2017:214-223.GulrajaniI,AhmedF,ArjovskyM,etal.Improvedtrainingof