基于去噪非整数位置像素图像重建方法研究.pdfVIP

DENOISING-BASEDIMAGERECONSTRUCTIONFROMPIXELSLOCATEDAT

NON-INTEGERPOSITIONS

Ja´nKoloda,Ju¨rgenSeilerandAndre´Kaup

ChairofMultimediaCommunicationsandSignalProcessing

Friedrich-AlexanderUniversityErlangen-Nu¨rnberg(FAU),Germany

Digitalimagesarecommonlyrepresentedasregular2Dar-

rays,sopixelsareorganizedinformofamatrixaddressedby

integers.However,therearemanyimageprocessingopera-

tions,suchasrotationormotioncompensation,thatproduce

pixelsatnon-integerpositions.Typically,imagereconstruc-

tiontechniquescannothandlesampatnon-integerposi-

tions.Inthispaper,weproposetousetriangulation-basedre-(a)(b)

constructionasinitialestimatethatislaterrefinedbyanovel

Fig.1:Exampleofapplicationthatyieldspixelsatnon-integerpo-

adaptivedenoisingframework.Simulationsrevealthatim-

sitions.Thesampin(b)areobtainedbyrotatingthepixelsin(a)

provementsofuptomorethan1.8dB(intermsofPSNR)are

by10degrees.Dashedlinesrepresenttheregulargrid.

achievedwithrespecttotheinitialestimate.

IndexTerms—Floatingmesh,adaptivedenoising,image

reconstructioncaseofalgorithmsbasedonexemrs[6,7]orautorregres-

sivemodelling[8].Reconstructionmethodscarried