Home // MMEDIA 2013, The Fifth International Conferences on Advances in Multimedia // View article

News Video Semantic Topic Mining Based on Multi-wing Harmoniums Model

Authors:
Xin Wen Xu
Yu Bo Shen
Guo Hui Li

Keywords: news video multi-wing Harmoniums; video mining; semantic mining; news video

Abstract:
Two-layer undirected graphical model, Harmoniums, is a new approach to mine latent semantic topics from observed data. For the multi-modal heterogeneous features of news video, this paper proposes multi-wing Harmoniums (MWH) model that represents news video stories as latent semantic topics derived by jointly modeling the transcript text, color histogram and edge histogram of the video. This model includes a multivariate Poisson distribution and two multivariate Gaussian distributions. It extends and improves earlier models based on two-layer random fields, which capture bidirectional dependencies between hidden topic aspects and observed inputs. The model especially facilitates efficient inference and robust topic mixing, and provides high flexibilities in modeling the latent topic spaces. The variational algorithm efficiently reduces the difficulty of model learning. The experiments results on the CCTV news video collections and an extensive comparison with various extant models show the efficiency of MWH on news video semantic mining.

Pages: 74 to 81

Copyright: Copyright (c) IARIA, 2013

Publication date: April 21, 2013

Published in: conference

ISSN: 2308-4448

ISBN: 978-1-61208-265-3

Location: Venice, Italy

Dates: from April 21, 2013 to April 26, 2013