image inpaintingreadmeWord文档下载推荐.docx

image inpaintingreadmeWord文档下载推荐.docx

《image inpaintingreadmeWord文档下载推荐.docx》由会员分享，可在线阅读，更多相关《image inpaintingreadmeWord文档下载推荐.docx（4页珍藏版）》请在冰豆网上搜索。

thepicturewindowandtheconcolewindow.Theprimaryuseractivityisdrawingthemaskovertheareatoberepaired.Themaskisdrawnwiththeleftmousebuttonanderasedwiththerightmousebutton.WhenevertheuserinitiatesafileI/Ooperation,thefilenameistobetypedintotheconsolewindow.

leftmousedrag-drawmask

rightmousedrag-erasemask

'

C'

-clearmask

u'

-unlimitedundo（ofmaskdrawing）

r'

-redo（maskdrawing）

U'

-toggleundoon/off（offsavessomememory--notmuch）

b'

-brushsize（rotatebetweenthreesizes）

t'

-togglemaskdisplayon/off

c'

-cyclethroughmaskcolors

s'

-saveimage

M'

-savemask

L'

-loadmask

+'

-oneiterationofinpainting

d'

-oneiterationofanisotropicdiffusion

D'

-togglediffusionon/off

i'

-100iterations,followedbyastepofdiffusion,ifdiffusionison

I'

-2000iterations,followedbyastepofdiffusion,ifdiffusionison

V'

-specifyhowmanyblocksof100iterationsofinpaintingtodo,followedbyastepof

diffusion,ifdiffusionison

a'

-15iterationsofinpainting,followedby2stepsofdiffusion

A'

-specifyhowmanyblocksof15i/2d

INPAINTING:

Theinpaintingalgorithmmeasuresthesmoothnessoftheimageandusesthistodecidehowtoextendcontoursthroughthemaskedregion.ItusestheLaplacianasasmoothnessindicator.To

findedgecontours,weusavectorperpendiculartothegradient.Thegradientoftheimagepointsinthedirectionofgreatestincrease.Alongedges,thegradientpointsperpendiculartotheedgebecausethegreatestchangeinthevalueoftheimageisacrossanedge.

Imageinpaintingisaniterativealgorithmthatonlychangespixelsinthemask.Inordertogetfromstepntostepn+1:

defineL（x,y）=2DLaplacianatpixel（x,y）

dLisavector,thechangeinsmoothnesstotheleft,right,top,andbottomofthecurrentpixel

Nisavector,avectorperpendiculartothegradient

Gisascalar,ameasureofthelengthofthegradient

Updateisapixelarray

Imageisapixelarray,theimageatstepn

ImageNextisapixelarray,theimageatstepn+1

deltatisascalar,thestepsizeforupdatinganimage

foreverypixel（x,y）inthemask

dL（x,y）=（L（x+1,y）-L（x-1,y）,L（x,y+1）-L（x,j-1））

N=normalizedvectorperpendiculartogradient

B=dL.N

G=length*（N）

Update（x,y）=B*G

ImageNext（x,y）=Image（x,y）+deltat*Update（x,y）

length*isaspecialmeasureoflengththatneedstobeusedinordertogetstableresults.Itis

definedas

_x-xdirection

_y-ydirection

_f-forwardderivative-calculatederivativeasdifferencebetweenpixelinfrontandcurrent

pixel

_b-backardderivative-calculatederivativeasdifferencebetweencurrentpixelandpixelin

back

_m-minimumofvalueand0

_M-maximumofvalueand0

if（B>

0）

length*=sqrt（Image_xbm^2+Image_xfM^2+Image_ybm^2+Image_yfM^2）

else

length*=sqrt（Image_xbM^2+Image_xfm^2+Image_ybM^2+Image_yfm^2）

ANISOTROPICDIFFUSION:

Aftereveryseveralstepsofinpainting,werunaniterationofanisotropicdiffusion.Anisotropic

diffusionisoneapproachtonoiseremovalbecauseittendstosharpenedgeswhilesmoothingover

continuousregions.Itisusedherefortworeasons.Ithelpstosmoothoversomeoftheparts

filledinbytheinpaintingalgorithm.Italsohelpstojoinedgesacrossthemaskastheyare

beingdrawninsothattheywillbecontinuous.WeadaptedRobertEstes'

s

MCDalgorithmforouruse.

OPTIMIZATIONS

Weimplementedanumberofoptimizationsinanefforttomaketheprogramfasterandmoreresponsive.TheimageiskeptinaformatthatGLcanreaddirectly（whichallowsforfairlyquickredraws）.However,forlargeimages,eventhisincursanoticabledelay,soformostmaskandimagemanipulations,wedrawthechangedpixelsdirectlytothescreentoavoidwhole-imageredraws.Themaskitselfisstoredinabitfield,andwedevelopedfastalgorithmsforiterationandmodificationwhichutilizethefactthatthemaskisusuallysparse.

Undoandredoarealsomemory-efficient;

althoughwemustdynamicallyallocatememoryastheuserinteractswiththeprogram,weuseadynamic,scalablevectorclasswhichavoidscostlylinkedlistsandpointers.Asaresulteachpixelchangerequiresonlyabout10bytesofstorageonaverage（4eachfor

xandycoords,1forindicatingthetypeofoperation,and1[onaverage]forpointingtothenextvector）.

Fortheinpaintinganddiffusionalgorithms,weusedinlinefunctionsandpass-by-referenceasmuchaspossible,andtriedtoavoidtediousiterationsoverthewholeimage.Diffusionisaverycostlyalgorithm;

ourcurrentimplementationrequiresabout3xasmuchmemorytorunastheoriginalimage.Therearelikelymoreefficientimplementationsoutthere,butwecouldn'

tthinkofabetterwaytoimplementitthatwouldstillbefast.

--------------------------------

PublishedPaper

psfile

OriginalPaper:

Preprint

http:

//www.ima.umn.edu/preprints/dec99/dec99.html

AnisotropicDiffusionCode

//info.cipic.ucdavis.edu/ftp/estes/

Implementationofinpainting

//www-users.itlabs.umn.edu/~rams0053/