Global stereo reconstruction under second-order smoothness priors

Second-order priors on the smoothness of 3D surfaces are a better model of typical scenes than first-order priors. However, stereo reconstruction using global inference algorithms, such as graph cuts, has not been able to incorporate second-order priors because the triple cliques needed to express t...

Full description

Bibliographic Details
Main Authors: Woodford, O, Torr, P, Reid, I, Fitzgibbon, A
Format: Journal article
Language:English
Published: IEEE 2009
_version_ 1826315257499353088
author Woodford, O
Torr, P
Reid, I
Fitzgibbon, A
author_facet Woodford, O
Torr, P
Reid, I
Fitzgibbon, A
author_sort Woodford, O
collection OXFORD
description Second-order priors on the smoothness of 3D surfaces are a better model of typical scenes than first-order priors. However, stereo reconstruction using global inference algorithms, such as graph cuts, has not been able to incorporate second-order priors because the triple cliques needed to express them yield intractable (nonsubmodular) optimization problems. This paper shows that inference with triple cliques can be effectively performed. Our optimization strategy is a development of recent extensions to α-expansion, based on the “QPBO” algorithm. The strategy is to repeatedly merge proposal depth maps using a novel extension of QPBO. Proposal depth maps can come from any source, for example, frontoparallel planes as in α-expansion, or indeed any existing stereo algorithm, with arbitrary parameter settings.
first_indexed 2024-12-09T03:22:39Z
format Journal article
id oxford-uuid:9af714b4-ce28-4142-9baf-7bfee4afc153
institution University of Oxford
language English
last_indexed 2024-12-09T03:22:39Z
publishDate 2009
publisher IEEE
record_format dspace
spelling oxford-uuid:9af714b4-ce28-4142-9baf-7bfee4afc1532024-11-19T14:46:27ZGlobal stereo reconstruction under second-order smoothness priorsJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:9af714b4-ce28-4142-9baf-7bfee4afc153EnglishSymplectic ElementsIEEE2009Woodford, OTorr, PReid, IFitzgibbon, ASecond-order priors on the smoothness of 3D surfaces are a better model of typical scenes than first-order priors. However, stereo reconstruction using global inference algorithms, such as graph cuts, has not been able to incorporate second-order priors because the triple cliques needed to express them yield intractable (nonsubmodular) optimization problems. This paper shows that inference with triple cliques can be effectively performed. Our optimization strategy is a development of recent extensions to α-expansion, based on the “QPBO” algorithm. The strategy is to repeatedly merge proposal depth maps using a novel extension of QPBO. Proposal depth maps can come from any source, for example, frontoparallel planes as in α-expansion, or indeed any existing stereo algorithm, with arbitrary parameter settings.
spellingShingle Woodford, O
Torr, P
Reid, I
Fitzgibbon, A
Global stereo reconstruction under second-order smoothness priors
title Global stereo reconstruction under second-order smoothness priors
title_full Global stereo reconstruction under second-order smoothness priors
title_fullStr Global stereo reconstruction under second-order smoothness priors
title_full_unstemmed Global stereo reconstruction under second-order smoothness priors
title_short Global stereo reconstruction under second-order smoothness priors
title_sort global stereo reconstruction under second order smoothness priors
work_keys_str_mv AT woodfordo globalstereoreconstructionundersecondordersmoothnesspriors
AT torrp globalstereoreconstructionundersecondordersmoothnesspriors
AT reidi globalstereoreconstructionundersecondordersmoothnesspriors
AT fitzgibbona globalstereoreconstructionundersecondordersmoothnesspriors