Geometric and Texture Inpainting by Gibbs Sampling

David Karl John Gustafsson; Kim Steenstrup Pedersen; Mads Nielsen

Geometric and Texture Inpainting by Gibbs Sampling

David Karl John Gustafsson, Kim Steenstrup Pedersen, Mads Nielsen

Abstract

This paper discuss a method suitable for inpainting
both large scale geometric structures and more
stochastic texture components. Image inpainting
concerns the problem of reconstructing the intensity
contents inside regions of missing data. Common
techniques for solving this problem are methods
based on variational calculus and based on
statistical methods. Variationalmethods are good at
reconstructing large scale geometric structures but
have a tendency to smooth away texture. On the
contrary statistical methods can reproduce texture
faithfully but fails to reconstruct large scale
structures. In this paper we use the well-known
FRAME (Filters, Random Fields and Maximum Entropy)
for inpainting. We introduce a temperature term in
the learned FRAME Gibbs distribution. By sampling
using different temperature in the FRAME Gibbs
distribution, different contents of the image are
reconstructed. We propose a two step method for
inpainting using FRAME. First the geometric
structure of the image is reconstructed by sampling
from a cooled Gibbs distribution, then the
stochastic component is reconstructed by sample
froma heated Gibbs distribution. Both steps in the
reconstruction process are necessary, and contribute
in two very different ways to the appearance of the
reconstruction.

Original language	English
Title of host publication	Proceedings SSBA 2007 : Symposium on inage analysis, Linköping, March 14-14, 2007
Editors	Magnus Borga, Anders Brun, Michael Felsberg
Number of pages	4
Publisher	Linköpings Universitet
Publication date	2007
Publication status	Published - 2007
Event	Symposium on Image Analysis - Linköping, Sweden Duration: 14 Mar 2007 → 15 Mar 2007

Conference

Conference	Symposium on Image Analysis
Country/Territory	Sweden
City	Linköping
Period	14/03/2007 → 15/03/2007

Series	Institutionen för medicinsk teknik, Universitetet i Linköping
Number	LiU-IMT-R-0047
ISSN	0280-2813

Cite this

Geometric and Texture Inpainting by Gibbs Sampling. / Gustafsson, David Karl John; Pedersen, Kim Steenstrup ; Nielsen, Mads.
Proceedings SSBA 2007: Symposium on inage analysis, Linköping, March 14-14, 2007. ed. / Magnus Borga; Anders Brun; Michael Felsberg. Linköpings Universitet, 2007. (Institutionen för medicinsk teknik, Universitetet i Linköping; No. LiU-IMT-R-0047).

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Gustafsson, DKJ, Pedersen, KS & Nielsen, M 2007, Geometric and Texture Inpainting by Gibbs Sampling. in M Borga, A Brun & M Felsberg (eds), Proceedings SSBA 2007: Symposium on inage analysis, Linköping, March 14-14, 2007. Linköpings Universitet, Institutionen för medicinsk teknik, Universitetet i Linköping, no. LiU-IMT-R-0047, Symposium on Image Analysis, Linköping, Sweden, 14/03/2007.

Gustafsson, David Karl John ; Pedersen, Kim Steenstrup ; Nielsen, Mads. / Geometric and Texture Inpainting by Gibbs Sampling. Proceedings SSBA 2007: Symposium on inage analysis, Linköping, March 14-14, 2007. editor / Magnus Borga ; Anders Brun ; Michael Felsberg. Linköpings Universitet, 2007. (Institutionen för medicinsk teknik, Universitetet i Linköping; No. LiU-IMT-R-0047).

@inproceedings{f82eef80b55711dcbee902004c4f4f50,

title = "Geometric and Texture Inpainting by Gibbs Sampling",

abstract = "This paper discuss a method suitable for inpaintingboth large scale geometric structures and more stochastic texture components. Image inpaintingconcerns the problem of reconstructing the intensitycontents inside regions of missing data. Commontechniques for solving this problem are methods based on variational calculus and based onstatistical methods. Variationalmethods are good atreconstructing large scale geometric structures buthave a tendency to smooth away texture. On the contrary statistical methods can reproduce texturefaithfully but fails to reconstruct large scalestructures. In this paper we use the well-knownFRAME (Filters, Random Fields and Maximum Entropy) for inpainting. We introduce a temperature term inthe learned FRAME Gibbs distribution. By samplingusing different temperature in the FRAME Gibbsdistribution, different contents of the image are reconstructed. We propose a two step method forinpainting using FRAME. First the geometricstructure of the image is reconstructed by samplingfrom a cooled Gibbs distribution, then the stochastic component is reconstructed by samplefroma heated Gibbs distribution. Both steps in thereconstruction process are necessary, and contributein two very different ways to the appearance of the reconstruction.",

author = "Gustafsson, {David Karl John} and Pedersen, {Kim Steenstrup} and Mads Nielsen",

year = "2007",

language = "English",

series = "Institutionen f{\"o}r medicinsk teknik, Universitetet i Link{\"o}ping",

publisher = "Link{\"o}pings Universitet",

number = "LiU-IMT-R-0047",

editor = "Magnus Borga and Anders Brun and Michael Felsberg",

booktitle = "Proceedings SSBA 2007",

note = "Symposium on Image Analysis ; Conference date: 14-03-2007 Through 15-03-2007",

}

TY - GEN

T1 - Geometric and Texture Inpainting by Gibbs Sampling

AU - Gustafsson, David Karl John

AU - Pedersen, Kim Steenstrup

AU - Nielsen, Mads

PY - 2007

Y1 - 2007

N2 - This paper discuss a method suitable for inpaintingboth large scale geometric structures and more stochastic texture components. Image inpaintingconcerns the problem of reconstructing the intensitycontents inside regions of missing data. Commontechniques for solving this problem are methods based on variational calculus and based onstatistical methods. Variationalmethods are good atreconstructing large scale geometric structures buthave a tendency to smooth away texture. On the contrary statistical methods can reproduce texturefaithfully but fails to reconstruct large scalestructures. In this paper we use the well-knownFRAME (Filters, Random Fields and Maximum Entropy) for inpainting. We introduce a temperature term inthe learned FRAME Gibbs distribution. By samplingusing different temperature in the FRAME Gibbsdistribution, different contents of the image are reconstructed. We propose a two step method forinpainting using FRAME. First the geometricstructure of the image is reconstructed by samplingfrom a cooled Gibbs distribution, then the stochastic component is reconstructed by samplefroma heated Gibbs distribution. Both steps in thereconstruction process are necessary, and contributein two very different ways to the appearance of the reconstruction.

AB - This paper discuss a method suitable for inpaintingboth large scale geometric structures and more stochastic texture components. Image inpaintingconcerns the problem of reconstructing the intensitycontents inside regions of missing data. Commontechniques for solving this problem are methods based on variational calculus and based onstatistical methods. Variationalmethods are good atreconstructing large scale geometric structures buthave a tendency to smooth away texture. On the contrary statistical methods can reproduce texturefaithfully but fails to reconstruct large scalestructures. In this paper we use the well-knownFRAME (Filters, Random Fields and Maximum Entropy) for inpainting. We introduce a temperature term inthe learned FRAME Gibbs distribution. By samplingusing different temperature in the FRAME Gibbsdistribution, different contents of the image are reconstructed. We propose a two step method forinpainting using FRAME. First the geometricstructure of the image is reconstructed by samplingfrom a cooled Gibbs distribution, then the stochastic component is reconstructed by samplefroma heated Gibbs distribution. Both steps in thereconstruction process are necessary, and contributein two very different ways to the appearance of the reconstruction.

M3 - Article in proceedings

T3 - Institutionen för medicinsk teknik, Universitetet i Linköping

BT - Proceedings SSBA 2007

A2 - Borga, Magnus

A2 - Brun, Anders

A2 - Felsberg, Michael

PB - Linköpings Universitet

T2 - Symposium on Image Analysis

Y2 - 14 March 2007 through 15 March 2007

ER -

Geometric and Texture Inpainting by Gibbs Sampling

Abstract

Conference

Fingerprint

Cite this