UNDIRECTEDGRAPHICALMODELS FORVIDEOANALYSISAND CLASSIFICATION

Alexander G. Hauptmann

Abstract

Accurate andefficient videoclassification andretrieval demandsthefusion ofmultimodal information andtheuseof intermediate representations. Thispaperdescribes anundirected graphical modelbasedonexponential-family harmonium, whichderives intermediate semantic representations of video databyjointly modeling thetextual andimageinformation inthevideo. Wepropose anextension ofthemodel toderive category-specific video representation andintegrate video classification asapartofthemodeling process. We report satisfactory classification performance onasetof15 video categories fromTRECVIDcollection aswellascomparison ontheeffectiveness ofdifferent inference algorithms.
BibTeX
@inproceedings{hauptmann2007undirectedgraphicalmodels,
  author = {Alexander G. Hauptmann},
  title = {{UNDIRECTEDGRAPHICALMODELS FORVIDEOANALYSISAND CLASSIFICATION}},
  booktitle = {ICME},
  year = {2007},
  doi = {10.1109/icme.2007.4284945},
}

Loading related content...

Figure