DocumentCode :
3256762
Title :
Exact optimization conditions for discrete linear inverse problems
Author :
Tuysuzoglu, A. ; Yilmaz, Ender ; Karl, W.C. ; Castanon, David
Author_Institution :
Boston Univ., Boston, MA, USA
fYear :
2013
fDate :
3-5 Dec. 2013
Firstpage :
1117
Lastpage :
1121
Abstract :
Recently, graph cut methods have been used with great success on discrete-label problems occurring in computer vision. Unfortunately, the presence of linear image mappings prevents the use of these techniques in image deconvolution. This work aims to expand the application of the successful graph-cut framework to linear inverse problems and deconvolution. We analyze the structure of linear inverse problems, showing the relationship of the sensing structure to graph non-representability of the problem and use insights from our analysis to present a class of linear operators that is graph representable. We propose a new method of variable relabeling that can transform a class of non-representable problems of this type to corresponding ones which are graph representable, thus allowing the use of graph-cut techniques for these problems.
Keywords :
computer vision; deconvolution; graph theory; optimisation; computer vision; discrete linear inverse problems; discrete-label problems; exact optimization conditions; graph cut methods; image deconvolution; linear image mappings; linear operators; nonrepresentable problems; sensing structure; variable relabeling; Computed tomography; Computer vision; Deconvolution; Image reconstruction; Minimization; Noise measurement;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Global Conference on Signal and Information Processing (GlobalSIP), 2013 IEEE
Conference_Location :
Austin, TX
Type :
conf
DOI :
10.1109/GlobalSIP.2013.6737090
Filename :
6737090
Link To Document :
بازگشت