Presentation 2013-06-10
Topic Models Taking into Account Burstiness of Local Features in Video
Yang XIE, Koji EGUCHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper we propose a topic model, Corr-DCMLDA, which can integrate visual words and the corresponding speech transcript words for key-frame images in each video data. Corr-DCMLDA has two distinguishing characteristics. One is the generative process that it first generates topics for visual words and then only the topics associated with the visual words are used to generate speech transcript words. The other is the assumption that the same latent topics are represented with different visual words and speech transcript words in different videos. We demonstrate through experiments on genre classification with online videos that Corr-DCMLDA works more effectively than some prior models.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Video topic modeling / Dirichlet compound multinomial models / collapsed Gibbs sampling
Paper # PRMU2013-20
Date of Issue

Conference Information
Committee PRMU
Conference Date 2013/6/3(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Pattern Recognition and Media Understanding (PRMU)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Topic Models Taking into Account Burstiness of Local Features in Video
Sub Title (in English)
Keyword(1) Video topic modeling
Keyword(2) Dirichlet compound multinomial models
Keyword(3) collapsed Gibbs sampling
1st Author's Name Yang XIE
1st Author's Affiliation Graduate School of Systems Informatics, Kobe University()
2nd Author's Name Koji EGUCHI
2nd Author's Affiliation Graduate School of Systems Informatics, Kobe University
Date 2013-06-10
Paper # PRMU2013-20
Volume (vol) vol.113
Number (no) 75
Page pp.pp.-
#Pages 6
Date of Issue