You need JavaScript to view this

Bayesian methods to restore and re build images: application to gamma-graphy and to photofission tomography; Methodes bayesiennes pour la restauration et la reconstruction d`images application a la gammagraphie et a la tomographie par photofissions

Abstract

Bayesian algorithms are developed to solve inverse problems in gamma imaging and photofission tomography. The first part of this work is devoted to the modeling of our measurement systems. Two models have been found for both applications: the first one is a simple conventional model and the second one is a cascaded point process model. EM and MCMC Bayesian algorithms for image restoration and image reconstruction have been developed for these models and compared. The cascaded point process model does not improve significantly the results previously obtained by the classical model. To original approaches have been proposed, which increase the results previously obtained. The first approach uses an inhomogeneous Markov Random Field as a prior law, and makes the regularization parameter spatially vary. However, the problem of the estimation of hyper-parameters has not been solved. In the case of the deconvolution of point sources, a second approach has been proposed, which introduces a high level prior model. The picture is modeled as a list of objects, whose parameters and number are unknown. The results obtained with this method are more accurate than those obtained with the conventional Markov Random Field prior model and require less computational costs. (author)
Authors:
Publication Date:
Oct 26, 1998
Product Type:
Thesis/Dissertation
Report Number:
FRCEA-TH-737
Reference Number:
SCA: 550601; PA: AIX-30:044364; EDB-99:103184; SN: 99002152891
Resource Relation:
Other Information: TH: These Sciences; PBD: 26 Oct 1998
Subject:
55 BIOLOGY AND MEDICINE, BASIC STUDIES; GAMMA RADIOGRAPHY; IMAGE PROCESSING; IMAGES; INVERSE SCATTERING PROBLEM; MARKOV PROCESS; MONTE CARLO METHOD; TOMOGRAPHY
OSTI ID:
690292
Research Organizations:
CEA Grenoble, 38 (France); Paris-11 Univ., 91 - Orsay (France)
Country of Origin:
France
Language:
French
Other Identifying Numbers:
Other: ON: DE99635220; TRN: FR9905942044364
Availability:
INIS; OSTI as DE99635220
Submitting Site:
FRN
Size:
210 p.
Announcement Date:

Citation Formats

Stawinski, G. Bayesian methods to restore and re build images: application to gamma-graphy and to photofission tomography; Methodes bayesiennes pour la restauration et la reconstruction d`images application a la gammagraphie et a la tomographie par photofissions. France: N. p., 1998. Web.
Stawinski, G. Bayesian methods to restore and re build images: application to gamma-graphy and to photofission tomography; Methodes bayesiennes pour la restauration et la reconstruction d`images application a la gammagraphie et a la tomographie par photofissions. France.
Stawinski, G. 1998. "Bayesian methods to restore and re build images: application to gamma-graphy and to photofission tomography; Methodes bayesiennes pour la restauration et la reconstruction d`images application a la gammagraphie et a la tomographie par photofissions." France.
@misc{etde_690292,
title = {Bayesian methods to restore and re build images: application to gamma-graphy and to photofission tomography; Methodes bayesiennes pour la restauration et la reconstruction d`images application a la gammagraphie et a la tomographie par photofissions}
author = {Stawinski, G}
abstractNote = {Bayesian algorithms are developed to solve inverse problems in gamma imaging and photofission tomography. The first part of this work is devoted to the modeling of our measurement systems. Two models have been found for both applications: the first one is a simple conventional model and the second one is a cascaded point process model. EM and MCMC Bayesian algorithms for image restoration and image reconstruction have been developed for these models and compared. The cascaded point process model does not improve significantly the results previously obtained by the classical model. To original approaches have been proposed, which increase the results previously obtained. The first approach uses an inhomogeneous Markov Random Field as a prior law, and makes the regularization parameter spatially vary. However, the problem of the estimation of hyper-parameters has not been solved. In the case of the deconvolution of point sources, a second approach has been proposed, which introduces a high level prior model. The picture is modeled as a list of objects, whose parameters and number are unknown. The results obtained with this method are more accurate than those obtained with the conventional Markov Random Field prior model and require less computational costs. (author)}
place = {France}
year = {1998}
month = {Oct}
}