Correlated topic model for image annotation

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

For the task of image annotation, traditional methods based on probabilistic topic model, such as correspondence Latent Dirichlet Allocation (corrLDA) [1], assumes that image is a mixture of latent topics. However, this kind of models is unable to directly model correlation between topics since topic proportions of an image are generated independently. Our model, called correspondence Correlated Topic Model (corrCTM), extends Correlated Topic Model (CTM) [2] from natural language processing to capture topic correlation from covariance structure of more flexible model distribution. Unlike previous LDA based models, topic proportions are correlated with each other in proposed corrCTM. And the topic correlation propagates from image features to annotation words through a generative process, and finally correspondence between images and words could be generated. We derive an approximate inference and estimation algorithm based on variational method. We examine the performance of our model on two benchmark image datasets, show improved performance over corrLDA for both annotation and modeling word correlation.

Original languageEnglish
Title of host publicationFCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision
Pages201-208
Number of pages8
DOIs
Publication statusPublished - Apr 15 2013
Event19th Korea-Japan Joint Workshop on Frontiers of Computer Vision, FCV 2013 - Incheon, Korea, Republic of
Duration: Jan 30 2013Feb 1 2013

Other

Other19th Korea-Japan Joint Workshop on Frontiers of Computer Vision, FCV 2013
CountryKorea, Republic of
CityIncheon
Period1/30/132/1/13

Fingerprint

Processing
Statistical Models

All Science Journal Classification (ASJC) codes

  • Computer Vision and Pattern Recognition

Cite this

Xu, X., Shimada, A., & Taniguchi, R-I. (2013). Correlated topic model for image annotation. In FCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision (pp. 201-208). [6485488] https://doi.org/10.1109/FCV.2013.6485488

Correlated topic model for image annotation. / Xu, Xing; Shimada, Atsushi; Taniguchi, Rin-Ichiro.

FCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision. 2013. p. 201-208 6485488.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Xu, X, Shimada, A & Taniguchi, R-I 2013, Correlated topic model for image annotation. in FCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision., 6485488, pp. 201-208, 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision, FCV 2013, Incheon, Korea, Republic of, 1/30/13. https://doi.org/10.1109/FCV.2013.6485488
Xu X, Shimada A, Taniguchi R-I. Correlated topic model for image annotation. In FCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision. 2013. p. 201-208. 6485488 https://doi.org/10.1109/FCV.2013.6485488
Xu, Xing ; Shimada, Atsushi ; Taniguchi, Rin-Ichiro. / Correlated topic model for image annotation. FCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision. 2013. pp. 201-208
@inproceedings{ec366ce1095948978a7d1c95949f4886,
title = "Correlated topic model for image annotation",
abstract = "For the task of image annotation, traditional methods based on probabilistic topic model, such as correspondence Latent Dirichlet Allocation (corrLDA) [1], assumes that image is a mixture of latent topics. However, this kind of models is unable to directly model correlation between topics since topic proportions of an image are generated independently. Our model, called correspondence Correlated Topic Model (corrCTM), extends Correlated Topic Model (CTM) [2] from natural language processing to capture topic correlation from covariance structure of more flexible model distribution. Unlike previous LDA based models, topic proportions are correlated with each other in proposed corrCTM. And the topic correlation propagates from image features to annotation words through a generative process, and finally correspondence between images and words could be generated. We derive an approximate inference and estimation algorithm based on variational method. We examine the performance of our model on two benchmark image datasets, show improved performance over corrLDA for both annotation and modeling word correlation.",
author = "Xing Xu and Atsushi Shimada and Rin-Ichiro Taniguchi",
year = "2013",
month = "4",
day = "15",
doi = "10.1109/FCV.2013.6485488",
language = "English",
isbn = "9781467356206",
pages = "201--208",
booktitle = "FCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision",

}

TY - GEN

T1 - Correlated topic model for image annotation

AU - Xu, Xing

AU - Shimada, Atsushi

AU - Taniguchi, Rin-Ichiro

PY - 2013/4/15

Y1 - 2013/4/15

N2 - For the task of image annotation, traditional methods based on probabilistic topic model, such as correspondence Latent Dirichlet Allocation (corrLDA) [1], assumes that image is a mixture of latent topics. However, this kind of models is unable to directly model correlation between topics since topic proportions of an image are generated independently. Our model, called correspondence Correlated Topic Model (corrCTM), extends Correlated Topic Model (CTM) [2] from natural language processing to capture topic correlation from covariance structure of more flexible model distribution. Unlike previous LDA based models, topic proportions are correlated with each other in proposed corrCTM. And the topic correlation propagates from image features to annotation words through a generative process, and finally correspondence between images and words could be generated. We derive an approximate inference and estimation algorithm based on variational method. We examine the performance of our model on two benchmark image datasets, show improved performance over corrLDA for both annotation and modeling word correlation.

AB - For the task of image annotation, traditional methods based on probabilistic topic model, such as correspondence Latent Dirichlet Allocation (corrLDA) [1], assumes that image is a mixture of latent topics. However, this kind of models is unable to directly model correlation between topics since topic proportions of an image are generated independently. Our model, called correspondence Correlated Topic Model (corrCTM), extends Correlated Topic Model (CTM) [2] from natural language processing to capture topic correlation from covariance structure of more flexible model distribution. Unlike previous LDA based models, topic proportions are correlated with each other in proposed corrCTM. And the topic correlation propagates from image features to annotation words through a generative process, and finally correspondence between images and words could be generated. We derive an approximate inference and estimation algorithm based on variational method. We examine the performance of our model on two benchmark image datasets, show improved performance over corrLDA for both annotation and modeling word correlation.

UR - http://www.scopus.com/inward/record.url?scp=84875982111&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84875982111&partnerID=8YFLogxK

U2 - 10.1109/FCV.2013.6485488

DO - 10.1109/FCV.2013.6485488

M3 - Conference contribution

AN - SCOPUS:84875982111

SN - 9781467356206

SP - 201

EP - 208

BT - FCV 2013 - Proceedings of the 19th Korea-Japan Joint Workshop on Frontiers of Computer Vision

ER -