2024年4月21日发(作者：联想k900评测)

Anon-localalgorithmforimagedenoising

AntoniBuades,BartomeuColl

`atiquesiInform`atica,UIB

mossaKm.7.5,

07122PalmadeMallorca,Spain

vdmiabc4@,@

Abstract

Weproposeanewmeasure,themethodnoise,toevalu-

ateandcomparetheperformanceofdigitalimagedenois-

firstcomputeandanalyzethismethod

noiseforawideclassofdenoisingalgorithms,namelythe

localsmoothingfi,weproposeanewalgo-

rithm,thenonlocalmeans(NL-means),basedonanonlo-

y,wepresent

someexperimentscomparingtheNL-meansalgorithmand

thelocalsmoothingfilters.

Jean-MichelMorel

CMLA,ENSCachan

61,AvduPr´esidentWilson

94235Cachan,France

morel@

FormallywedefineadenoisingmethodD

h

asadecom-

position

v=D

h

v+n(D

h

,v),

wherevisthenoisyimageandhisafilteringparame-

terwhichusuallydependsonthestandarddeviationofthe

y,D

h

vissmootherthanvandn(D

h

,v)looks

omposition

ofanimagebetweenasmoothpartandanonsmoothor

oscillatorypartisacurrentsubjectofresearch(forexample

Osheretal.[10]).In[8],tudiedthesuitablefunc-

maryscopeof

thislatterstudyisnotdenoisingsincetheoscillatorypart

containsbothnoiseandtexture.

Thedenoisingmethodsshouldnotaltertheoriginalim-

,mostdenoisingmethodsdegradeorremovethe

firtobetterunderstand

thisremoval,weshallintroduceandanalyzethemethod

hodnoiseisdefinedasthedifferencebe-

tweentheoriginal(alwaysslightlynoisy)imageuandits

denoisedversion.

WealsoproposeandanalyzetheNL-meansalgorithm,

whichisdefinedbythesimpleformula