当前位置： 文档下载 > 所有分类 > Texture synthesis by non-parametric samplingA non-parametric method for texture synthesis is proposed. The texture synthesis process grows a new image outward from an initial seed, one pixel at a time. A Markov random field model is assumed, and the conditional distribution of a pixel given all its
IEEEInternationalConferenceonComputerVision,Corfu,Greece,September1999
TextureSynthesisbyNon-parametricSampling
AlexeiA.EfrosandThomasK.Leung
ComputerScienceDivisionUniversityofCalifornia,BerkeleyBerkeley,CA94720-1776,U.S.A.efros,leungt@cs.berkeley.edu
Abstract
Anon-parametricmethodfortexturesynthesisispro-posed.Thetexturesynthesisprocessgrowsanewimageoutwardfromaninitialseed,onepixelatatime.AMarkovrandom eldmodelisassumed,andtheconditionaldistri-butionofapixelgivenallitsneighborssynthesizedsofarisestimatedbyqueryingthesampleimageand ndingallsim-ilarneighborhoods.Thedegreeofrandomnessiscontrolledbyasingleperceptuallyintuitiveparameter.Themethodaimsatpreservingasmuchlocalstructureaspossibleandproducesgoodresultsforawidevarietyofsyntheticandreal-worldtextures.
ofspatiallocality.Theresultisaverysimpletexturesyn-thesisalgorithmthatworkswellonawiderangeoftexturesandisespeciallywell-suitedforconstrainedsynthesisprob-lems(hole- lling).
1.1.Previouswork
Mostrecentapproacheshaveposedtexturesynthesisinastatisticalsettingasaproblemofsamplingfromaprobabil-itydistribution.Zhuet.al.[12]modeltextureasaMarkovRandomFieldanduseGibbssamplingforsynthesis.Un-fortunately,Gibbssamplingisnotoriouslyslowandinfactitisnotpossibletoassesswhenithasconverged.HeegerandBergen[6]trytocoercearandomnoiseimageintoatexturesamplebymatchingthe lterresponsehistogramsatdifferentspatialscales.Whilethistechniqueworkswellonhighlystochastictextures,thehistogramsarenotpow-erfulenoughtorepresentmorestructuredtexturepatternssuchasbricks.
DeBonet[1]alsousesamulti-resolution lter-basedap-proachinwhichatexturepatchata nerscaleiscondi-tionedonits“parents”atthecoarserscales.Thealgorithmworksbytakingtheinputtexturesampleandrandomizingitinsuchawayastopreservetheseinter-scaledependen-cies.Thismethodcansuccessfullysynthesizeawiderangeoftexturesalthoughtherandomnessparameterseemstoex-hibitperceptuallycorrectbehavioronlyonlargelystochas-tictextures.Anotherdrawbackofthismethodisthewaytextureimageslargerthantheinputaregenerated.Thein-puttexturesampleissimplyreplicatedto llthedesireddi-mensionsbeforethesynthesisprocess,implicitlyassumingthatalltexturesaretilablewhichisclearlynotcorrect.ThelatestworkintexturesynthesisbySimoncelliandPortilla[9,11]isbasedon rstandsecondorderpropertiesofjointwaveletcoef cientsandprovidesimpressiveresults.Itcancapturebothstochasticandrepeatedtexturesquitewell,butstillfailstoreproducehighfrequencyinformationonsomehighlystructuredpatterns.
1.Introduction
Texturesynthesishasbeenanactiveresearchtopicincomputervisionbothasawaytoverifytextureanalysismethods,aswellasinitsownright.Potentialapplicationsofasuccessfultexturesynthesisalgorithmarebroad,in-cludingocclusion ll-in,lossyimageandvideocompres-sion,foregroundremoval,etc.
Theproblemoftexturesynthesiscanbeformulatedasfollows:letusde netextureassomevisualpatternonanin nite2-Dplanewhich,atsomescale,hasastationarydistribution.Givena nitesamplefromsometexture(animage),thegoalistosynthesizeothersamplesfromthesametexture.Withoutadditionalassumptionsthisproblemisclearlyill-posedsinceagiventexturesamplecouldhavebeendrawnfromanin nitenumberofdifferenttextures.Theusualassumptionisthatthesampleislargeenoughthatitsomehowcapturesthestationarityofthetextureandthatthe(approximate)scaleofthetextureelements(texels)isknown.
Textureshavebeentraditionallyclassi edaseitherreg-ular(consistingofrepeatedtexels)orstochastic(withoutexplicittexels).However,almostallreal-worldtexturesliesomewhereinbetweenthesetwoextremesandshouldbecapturedwithasinglemodel.Inthispaperwehavechosenastatisticalnon-parametricmodelbasedontheassumption
Texture Synthesis by Non-Parametric Sampling ? Efros and Leung ? ICCV 1999 Texture Synthesis ? Fast Texture Synthesis using Tree-Structured Vector ...
Pixel-based Texture Synthesis 3. 4. 5. 6. A.A. Efros and T.K. Leung, “Texture Synthesis by Non-parametric Sampling”, ICCV, 1998. L. Wei and...
1 Introduction Practical Eye Movement Model using Texture Synthesis_专业资料...1999, Texture Synthesis by Non-parametric Sampling, ICCV’99, 1033—1038. ...
Leung, “Texture synthesis by non-parametric sampling,” in iccv. Published by the IEEE Computer Society, 1999, p. 1033. [2] A. Efros, W. Freeman...
Image quilting for texture synthesis and transfer_专业资料。The Problem: A...K. Leung. Texture synthesis by non-parametric sampling. In International ...
[12] is very complex, inspired by a non-parametric sampling method successfully used in texture synthesis [4], we construct the discrete probability ...
(10):1—62. 14 J Efros A,LeungThomas K.Texture synthesis by non—parametric sampling lJ J.Internationla Conference on Computer Vision,2005,8 (2):...
A. Efros and T. Leung. Texture synthesis by non-parametric sampling. In Proc. ICCV, pp. 1033–1038, Kerkyra, Greece, Sep 1999. ? ? ...
Various approaches towards dynamic texture synthesis can be classi?ed into nonparametric and parametric methods. The non-parametric methods directly sample the ...
iii APPLICATION OF NON-PARAMETRIC TEXTURE SYNTHESIS TO IMAGE INPAINTING BY ...The non-parametric sampling procedure for image inpainting utilizes a series ...
Much faster synthesis is obtained with non-parametric sampling [5], [6], [12]. Also assuming Markovianity of the texture, these techniques treat the ...
We build upon the work of [3] which uses non-parametric texture synthesis...Leung, “Texture synthesis by nonparametric sampling.”, Proceedings of the ...
我要评论