基于深度學(xué)習(xí)的圖像超分辨率重建研究

1、畢業(yè)設(shè)計(論文)基于深度學(xué)習(xí)的圖像超分辨率重建研究院別數(shù)學(xué)與統(tǒng)計學(xué)院專業(yè)名稱信息與計算科學(xué)班級學(xué)號學(xué)生姓名楚文玉指導(dǎo)教師張琨2017年06月10日東北大學(xué)秦皇島分校畢業(yè)設(shè)計(論文)第II頁ImageSuperResolutionUsingDeeplearningAuth:ChuWenyuTut:ZhangKunAbstractArtificialNeuralwkbecauseofitsstrongabilitytolearngetrap

2、iddevelopmentofartificialintelligencelettheArtificialNeuralwkbecometheresearchupsurgeagain.Deeplearninghasbeenwidelyusedincomputervisionspeechprocessingnaturallanguageprocessingsoon.Thesuperresolution(SR)techniqueisdesig

3、nedtorefactalowresolutionimagethroughaseriesofalgithmstoreconstructthecrespondinghighresolutionimage.CurrentlythemethodoffrequencydomainNonunifmimageinterpolationProjectionontoconvexset(POCS)Maximumaposteri(MPA)sparsemat

4、rixmethodarethemematuremethods.Thispapermainlyresearchestherealizationofsuperresolution(SR)reconstructionusingdeeplearning.Inthisthesisfirstisabriefintroductionofthedevelopmentofartificialneuralwkthenintroducestheapplica

5、tionofdeeplearningincomputervision.Withthatintroducessometheeticalknowledgeofneuralwkfinallyintroducestheconvolutionneuralwk(CNN)indeeplearning.Thisarticlemainlyresearcheshowtousetheconvolutionneuralwk(CNN)togetthesuperr

6、esolutionreconstruction.TheconvolutionneuralwkcontainsthreestructurestheeffectofthefirstlayerisPatchextractionrepresentationthesecondisthefunctionofNonlinearmappingtheroleofthethirdlayeristhehighresolutionimagereconstruc

7、tion.Firsttodownscalebicubicinterpolationanimageasthelowresolutionimagesastheinputoftheconvolutionneuralwkthehighresolutionimageastheoutputoftheconvolutionneuralwkusingconvolutionneuralwkestablishedendtoendmappingbetween