31 semi-coupled dictionary learning with applications image super resolution and photo sketch synthesis31半耦合词典学习并应用于图像超分辨率照片佐剂合成.pdf
Semi-CoupledDictionaryLearningwithApplicationstoImageSuper-Resolution
andPhoto-SketchSynthesis
ShenlongWangLeiZhangYanLiangQuanPan
NorthwesternPolytechnicalUniversity,TheHongKongPolytechnicUniversity
shenlong.wang@,cslzhang@.hk,liangyan@,quanpan@
Abstractderstandhowthehumanvisualsystemperceivesthedistinc-
tiveinformationofthesamesceneacrossdifferentsources.
Invariouscomputervisionapplications,oftenweneedInthepastdecades,imagecross-stylesynthesisand
toconvertanimageinonestyleintoanotherstyleforbet-recognitionhavebeenattractingmuchattention.Sincethe
tervisualization,interpretationandrecognition;forexam-imagesunderdifferentstyles,evendescribingthesame
ples,up-convertalowresolutionimagetoahighresolutionscene,canbeverydifferent,howtorevealtheunderlying
one,andconvertafacesketchintoaphotoformatching,relationsbetweenthetwostylesisthekeyissuetobestud-
etc.Asemi-coupleddictionarylearning(SCDL)modelisied.Inordertopredicttheunknownimagesinonestyle
proposedinthispapertosolvesuchcross-styleimagesyn-fromtheircounterpartsinanotherstyle,statisticallearning
thesisproblems.UnderSCDL,apairofdictionariesandaapproachescanbeadoptedtolearntheunderlyingmapping
mappingfunctionwillbesimultaneouslylearned.Thedic-fromexampleimagepairs.Manyimageprocessingand
tionarypaircanwellcharacterizethestructuraldomain-computervisiontasks