Multiscale Feature Extractors for Stereo Matching Cost Computation

Kyung Rae Kim, Yeong Jun Koh, Chang-Su Kim

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

We propose four efficient feature extractors based on convolutional neural networks for stereo matching cost computation. Two of them generate multiscale features with diverse receptive field sizes. These multiscale features are used to compute the corresponding multiscale matching costs. We then determine an optimal cost by combining the multiscale costs using edge information. On the other hand, the other two feature extractors produce uni-scale features by combining multiscale features directly through fully connected layers. Finally, after obtaining matching costs using one of the four extractors, we determine optimal disparities based on the cross-based cost aggregation and the semiglobal matching. Extensive experiments on the Middlebury stereo data sets demonstrate the effectiveness and efficiency of the proposed algorithm. Specifically, the proposed algorithm provides competitive matching performance with the state of the arts, while demanding lower computational complexity.

Original languageEnglish
Pages (from-to)27971-27983
Number of pages13
JournalIEEE Access
Volume6
DOIs
Publication statusPublished - 2018 May 17

Fingerprint

Costs
Computational complexity
Agglomeration
Neural networks
Experiments

Keywords

  • convolutional neural networks
  • matching cost computation
  • multiscale feature extraction
  • Stereo matching

ASJC Scopus subject areas

  • Computer Science(all)
  • Materials Science(all)
  • Engineering(all)

Cite this

Multiscale Feature Extractors for Stereo Matching Cost Computation. / Kim, Kyung Rae; Koh, Yeong Jun; Kim, Chang-Su.

In: IEEE Access, Vol. 6, 17.05.2018, p. 27971-27983.

Research output: Contribution to journalArticle

Kim, Kyung Rae ; Koh, Yeong Jun ; Kim, Chang-Su. / Multiscale Feature Extractors for Stereo Matching Cost Computation. In: IEEE Access. 2018 ; Vol. 6. pp. 27971-27983.
@article{816c77f532e74fc6b6e7a0032350412d,
title = "Multiscale Feature Extractors for Stereo Matching Cost Computation",
abstract = "We propose four efficient feature extractors based on convolutional neural networks for stereo matching cost computation. Two of them generate multiscale features with diverse receptive field sizes. These multiscale features are used to compute the corresponding multiscale matching costs. We then determine an optimal cost by combining the multiscale costs using edge information. On the other hand, the other two feature extractors produce uni-scale features by combining multiscale features directly through fully connected layers. Finally, after obtaining matching costs using one of the four extractors, we determine optimal disparities based on the cross-based cost aggregation and the semiglobal matching. Extensive experiments on the Middlebury stereo data sets demonstrate the effectiveness and efficiency of the proposed algorithm. Specifically, the proposed algorithm provides competitive matching performance with the state of the arts, while demanding lower computational complexity.",
keywords = "convolutional neural networks, matching cost computation, multiscale feature extraction, Stereo matching",
author = "Kim, {Kyung Rae} and Koh, {Yeong Jun} and Chang-Su Kim",
year = "2018",
month = "5",
day = "17",
doi = "10.1109/ACCESS.2018.2838442",
language = "English",
volume = "6",
pages = "27971--27983",
journal = "IEEE Access",
issn = "2169-3536",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Multiscale Feature Extractors for Stereo Matching Cost Computation

AU - Kim, Kyung Rae

AU - Koh, Yeong Jun

AU - Kim, Chang-Su

PY - 2018/5/17

Y1 - 2018/5/17

N2 - We propose four efficient feature extractors based on convolutional neural networks for stereo matching cost computation. Two of them generate multiscale features with diverse receptive field sizes. These multiscale features are used to compute the corresponding multiscale matching costs. We then determine an optimal cost by combining the multiscale costs using edge information. On the other hand, the other two feature extractors produce uni-scale features by combining multiscale features directly through fully connected layers. Finally, after obtaining matching costs using one of the four extractors, we determine optimal disparities based on the cross-based cost aggregation and the semiglobal matching. Extensive experiments on the Middlebury stereo data sets demonstrate the effectiveness and efficiency of the proposed algorithm. Specifically, the proposed algorithm provides competitive matching performance with the state of the arts, while demanding lower computational complexity.

AB - We propose four efficient feature extractors based on convolutional neural networks for stereo matching cost computation. Two of them generate multiscale features with diverse receptive field sizes. These multiscale features are used to compute the corresponding multiscale matching costs. We then determine an optimal cost by combining the multiscale costs using edge information. On the other hand, the other two feature extractors produce uni-scale features by combining multiscale features directly through fully connected layers. Finally, after obtaining matching costs using one of the four extractors, we determine optimal disparities based on the cross-based cost aggregation and the semiglobal matching. Extensive experiments on the Middlebury stereo data sets demonstrate the effectiveness and efficiency of the proposed algorithm. Specifically, the proposed algorithm provides competitive matching performance with the state of the arts, while demanding lower computational complexity.

KW - convolutional neural networks

KW - matching cost computation

KW - multiscale feature extraction

KW - Stereo matching

UR - http://www.scopus.com/inward/record.url?scp=85047227341&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85047227341&partnerID=8YFLogxK

U2 - 10.1109/ACCESS.2018.2838442

DO - 10.1109/ACCESS.2018.2838442

M3 - Article

AN - SCOPUS:85047227341

VL - 6

SP - 27971

EP - 27983

JO - IEEE Access

JF - IEEE Access

SN - 2169-3536

ER -