Nonlinear Regression of Saliency Guided Proposals for Unsupervised Segmentation of Dynamic Scenes

Yinhui ZHANG; Mohamed ABDEL-MOTTALEB; Zifen HE

doi:10.1587/transinf.2015EDP7295

Nonlinear Regression of Saliency Guided Proposals for Unsupervised Segmentation of Dynamic Scenes

Yinhui ZHANG, Mohamed ABDEL-MOTTALEB, Zifen HE

Full Text Views

0

Share
Cite this

Summary :

This paper proposes an efficient video object segmentation approach that is tolerant to complex scene dynamics. Unlike existing approaches that rely on estimating object-like proposals on an intra-frame basis, the proposed approach employs temporally consistent foreground hypothesis using nonlinear regression of saliency guided proposals across a video sequence. For this purpose, we first generate salient foreground proposals at superpixel level by leveraging a saliency signature in the discrete cosine transform domain. We propose to use a random forest based nonlinear regression scheme to learn both appearance and shape features from salient foreground regions in all frames of a sequence. Availability of such features can help rank every foreground proposals of a sequence, and we show that the regions with high ranking scores are well correlated with semantic foreground objects in dynamic scenes. Subsequently, we utilize a Markov Random Field to integrate both appearance and motion coherence of the top-ranked object proposals. A temporal nonlinear regressor for generating salient object support regions significantly improves the segmentation performance compared to using only per-frame objectness cues. Extensive experiments on challenging real-world video sequences are performed to validate the feasibility and superiority of the proposed approach for addressing dynamic scene segmentation.

Publication: IEICE TRANSACTIONS on Information Vol.E99-D No.2 pp.467-474

Publication Date: 2016/02/01

Publicized: 2015/11/06

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2015EDP7295

Type of Manuscript: PAPER

Category: Image Processing and Video Processing

Authors

Yinhui ZHANG
  Kunming University of Science and Technology
Mohamed ABDEL-MOTTALEB
  University of Miami,Effat University
Zifen HE
  Kunming University of Science and Technology

Keyword

video object segmentation, salient object-like proposal, nonlinear regressor, dynamic scene, random forest

Cite this

Copy

Yinhui ZHANG, Mohamed ABDEL-MOTTALEB, Zifen HE, "Nonlinear Regression of Saliency Guided Proposals for Unsupervised Segmentation of Dynamic Scenes" in IEICE TRANSACTIONS on Information, vol. E99-D, no. 2, pp. 467-474, February 2016, doi: 10.1587/transinf.2015EDP7295.
Abstract: This paper proposes an efficient video object segmentation approach that is tolerant to complex scene dynamics. Unlike existing approaches that rely on estimating object-like proposals on an intra-frame basis, the proposed approach employs temporally consistent foreground hypothesis using nonlinear regression of saliency guided proposals across a video sequence. For this purpose, we first generate salient foreground proposals at superpixel level by leveraging a saliency signature in the discrete cosine transform domain. We propose to use a random forest based nonlinear regression scheme to learn both appearance and shape features from salient foreground regions in all frames of a sequence. Availability of such features can help rank every foreground proposals of a sequence, and we show that the regions with high ranking scores are well correlated with semantic foreground objects in dynamic scenes. Subsequently, we utilize a Markov Random Field to integrate both appearance and motion coherence of the top-ranked object proposals. A temporal nonlinear regressor for generating salient object support regions significantly improves the segmentation performance compared to using only per-frame objectness cues. Extensive experiments on challenging real-world video sequences are performed to validate the feasibility and superiority of the proposed approach for addressing dynamic scene segmentation.
URL: https://globals.ieice.org/en_transactions/information/10.1587/transinf.2015EDP7295/_p

Copy

@ARTICLE{e99-d_2_467,
author={Yinhui ZHANG, Mohamed ABDEL-MOTTALEB, Zifen HE, },
journal={IEICE TRANSACTIONS on Information},
title={Nonlinear Regression of Saliency Guided Proposals for Unsupervised Segmentation of Dynamic Scenes},
year={2016},
volume={E99-D},
number={2},
pages={467-474},
abstract={This paper proposes an efficient video object segmentation approach that is tolerant to complex scene dynamics. Unlike existing approaches that rely on estimating object-like proposals on an intra-frame basis, the proposed approach employs temporally consistent foreground hypothesis using nonlinear regression of saliency guided proposals across a video sequence. For this purpose, we first generate salient foreground proposals at superpixel level by leveraging a saliency signature in the discrete cosine transform domain. We propose to use a random forest based nonlinear regression scheme to learn both appearance and shape features from salient foreground regions in all frames of a sequence. Availability of such features can help rank every foreground proposals of a sequence, and we show that the regions with high ranking scores are well correlated with semantic foreground objects in dynamic scenes. Subsequently, we utilize a Markov Random Field to integrate both appearance and motion coherence of the top-ranked object proposals. A temporal nonlinear regressor for generating salient object support regions significantly improves the segmentation performance compared to using only per-frame objectness cues. Extensive experiments on challenging real-world video sequences are performed to validate the feasibility and superiority of the proposed approach for addressing dynamic scene segmentation.},
keywords={},
doi={10.1587/transinf.2015EDP7295},
ISSN={1745-1361},
month={February},}

Copy

TY - JOUR
TI - Nonlinear Regression of Saliency Guided Proposals for Unsupervised Segmentation of Dynamic Scenes
T2 - IEICE TRANSACTIONS on Information
SP - 467
EP - 474
AU - Yinhui ZHANG
AU - Mohamed ABDEL-MOTTALEB
AU - Zifen HE
PY - 2016
DO - 10.1587/transinf.2015EDP7295
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E99-D
IS - 2
JA - IEICE TRANSACTIONS on Information
Y1 - February 2016
AB - This paper proposes an efficient video object segmentation approach that is tolerant to complex scene dynamics. Unlike existing approaches that rely on estimating object-like proposals on an intra-frame basis, the proposed approach employs temporally consistent foreground hypothesis using nonlinear regression of saliency guided proposals across a video sequence. For this purpose, we first generate salient foreground proposals at superpixel level by leveraging a saliency signature in the discrete cosine transform domain. We propose to use a random forest based nonlinear regression scheme to learn both appearance and shape features from salient foreground regions in all frames of a sequence. Availability of such features can help rank every foreground proposals of a sequence, and we show that the regions with high ranking scores are well correlated with semantic foreground objects in dynamic scenes. Subsequently, we utilize a Markov Random Field to integrate both appearance and motion coherence of the top-ranked object proposals. A temporal nonlinear regressor for generating salient object support regions significantly improves the segmentation performance compared to using only per-frame objectness cues. Extensive experiments on challenging real-world video sequences are performed to validate the feasibility and superiority of the proposed approach for addressing dynamic scene segmentation.
ER -