Skip to content
Back to outputs

CMIB: unsupervised image object categorization in multiple visual contexts

Research output: Contribution to journalArticle

Standard

CMIB: unsupervised image object categorization in multiple visual contexts. / Yan, Xiaoqiang; Ye, Yangdong; Qiu, Xueying; Manic, Milos; Yu, Hui.

In: IEEE Transactions on Industrial Informatics, Vol. 16, No. 6, 01.06.2020, p. 3974-3986.

Research output: Contribution to journalArticle

Harvard

Yan, X, Ye, Y, Qiu, X, Manic, M & Yu, H 2020, 'CMIB: unsupervised image object categorization in multiple visual contexts', IEEE Transactions on Industrial Informatics, vol. 16, no. 6, pp. 3974-3986. https://doi.org/10.1109/TII.2019.2939278

APA

Yan, X., Ye, Y., Qiu, X., Manic, M., & Yu, H. (2020). CMIB: unsupervised image object categorization in multiple visual contexts. IEEE Transactions on Industrial Informatics, 16(6), 3974-3986. https://doi.org/10.1109/TII.2019.2939278

Vancouver

Yan X, Ye Y, Qiu X, Manic M, Yu H. CMIB: unsupervised image object categorization in multiple visual contexts. IEEE Transactions on Industrial Informatics. 2020 Jun 1;16(6):3974-3986. https://doi.org/10.1109/TII.2019.2939278

Author

Yan, Xiaoqiang ; Ye, Yangdong ; Qiu, Xueying ; Manic, Milos ; Yu, Hui. / CMIB: unsupervised image object categorization in multiple visual contexts. In: IEEE Transactions on Industrial Informatics. 2020 ; Vol. 16, No. 6. pp. 3974-3986.

Bibtex

@article{f2fa019d4002483e82d187d7c59c28ea,
title = "CMIB: unsupervised image object categorization in multiple visual contexts",
abstract = "Object categorization in images is fundamental to various industrial areas, such as automated visual inspection, fast image retrieval and intelligent surveillance. Most existing methods treat visual features (e.g., scale-invariant feature transform, SIFT) as content information of the objects, while regarding image tags as its contextual information. However, the image tags can hardly been acquired in complete unsupervised settings, especially when the image volume is too large to be marked. In this work, we propose a novel and effective method called contextual multivariate information bottleneck (CMIB) to discover object category in totally unlabeled images. Unlike treating image tags as the object{\textquoteright}s context, CMIB adopts one feature representation of the images to characterize the object{\textquoteright}s content information, while regarding the auxiliary clusterings obtained by other multiple related features as its visual contexts. In the proposed CMIB framework, we borrow the idea of the data compression procedure for object category discovery, which aims to squeeze the source image collection into its compressed representation as much as possible, while maximally preserving the correlative information between the content and visual contexts. Specifically, two Bayesian networks are built to characterize the relationships between data compression and information preservation. Moreover, a sequential informationtheoretic optimization is proposed to ensure the convergence of the CMIB objective function. Extensive experiments on five real-world image data sets show that the proposed method can significantly outperform the state-of-the-art baselines.",
author = "Xiaoqiang Yan and Yangdong Ye and Xueying Qiu and Milos Manic and Hui Yu",
year = "2020",
month = jun,
day = "1",
doi = "10.1109/TII.2019.2939278",
language = "English",
volume = "16",
pages = "3974--3986",
journal = "IEEE Transactions on Industrial Informatics",
issn = "1551-3203",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "6",

}

RIS

TY - JOUR

T1 - CMIB: unsupervised image object categorization in multiple visual contexts

AU - Yan, Xiaoqiang

AU - Ye, Yangdong

AU - Qiu, Xueying

AU - Manic, Milos

AU - Yu, Hui

PY - 2020/6/1

Y1 - 2020/6/1

N2 - Object categorization in images is fundamental to various industrial areas, such as automated visual inspection, fast image retrieval and intelligent surveillance. Most existing methods treat visual features (e.g., scale-invariant feature transform, SIFT) as content information of the objects, while regarding image tags as its contextual information. However, the image tags can hardly been acquired in complete unsupervised settings, especially when the image volume is too large to be marked. In this work, we propose a novel and effective method called contextual multivariate information bottleneck (CMIB) to discover object category in totally unlabeled images. Unlike treating image tags as the object’s context, CMIB adopts one feature representation of the images to characterize the object’s content information, while regarding the auxiliary clusterings obtained by other multiple related features as its visual contexts. In the proposed CMIB framework, we borrow the idea of the data compression procedure for object category discovery, which aims to squeeze the source image collection into its compressed representation as much as possible, while maximally preserving the correlative information between the content and visual contexts. Specifically, two Bayesian networks are built to characterize the relationships between data compression and information preservation. Moreover, a sequential informationtheoretic optimization is proposed to ensure the convergence of the CMIB objective function. Extensive experiments on five real-world image data sets show that the proposed method can significantly outperform the state-of-the-art baselines.

AB - Object categorization in images is fundamental to various industrial areas, such as automated visual inspection, fast image retrieval and intelligent surveillance. Most existing methods treat visual features (e.g., scale-invariant feature transform, SIFT) as content information of the objects, while regarding image tags as its contextual information. However, the image tags can hardly been acquired in complete unsupervised settings, especially when the image volume is too large to be marked. In this work, we propose a novel and effective method called contextual multivariate information bottleneck (CMIB) to discover object category in totally unlabeled images. Unlike treating image tags as the object’s context, CMIB adopts one feature representation of the images to characterize the object’s content information, while regarding the auxiliary clusterings obtained by other multiple related features as its visual contexts. In the proposed CMIB framework, we borrow the idea of the data compression procedure for object category discovery, which aims to squeeze the source image collection into its compressed representation as much as possible, while maximally preserving the correlative information between the content and visual contexts. Specifically, two Bayesian networks are built to characterize the relationships between data compression and information preservation. Moreover, a sequential informationtheoretic optimization is proposed to ensure the convergence of the CMIB objective function. Extensive experiments on five real-world image data sets show that the proposed method can significantly outperform the state-of-the-art baselines.

U2 - 10.1109/TII.2019.2939278

DO - 10.1109/TII.2019.2939278

M3 - Article

VL - 16

SP - 3974

EP - 3986

JO - IEEE Transactions on Industrial Informatics

JF - IEEE Transactions on Industrial Informatics

SN - 1551-3203

IS - 6

ER -

ID: 15859476