TY - JOUR
T1 - Multimedia event detection with ℓ2-regularized logistic Gaussian mixture regression
AU - Liu, Changyu
AU - Dong, Shoubin
AU - Lu, Bin
AU - Abdel-Mottaleb, Mohamed
N1 - Funding Information:
This paper was supported by the National Natural Science Foundation of China under Grant No. 61070092, the Doctor Startup Foundation of Wuyi University under Grant No. 2014BS07, the Science and Technology Planning Project of Guangzhou under Grant No. 2013Y2-00073, and the Special Province-Ministry Funds for Industry, Education and Research of Guangdong Province under Grant No. 2013B090500087. We would like to thank anonymous reviewers for helpful comments.
Publisher Copyright:
© 2015, The Natural Computing Applications Forum.
PY - 2015/10/21
Y1 - 2015/10/21
N2 - Multimedia event detection (MED) is one of the most important branches of multimedia content analysis. Current research work on MED focuses mainly on detecting specific events, such as sport events, news events and suspicious events, which is far from achieving a complicated and generic MED due to the fact that these events usually contain a lot of visual attributes, such as objects, scenes and human actions. Being different from visual features, visual attributes are hidden classes to event detectors and event classifiers. Hence, proper representation of these visual attributes could be helpful in building a sophisticated and generic MED. In this paper, we use Gaussian mixture model (GMM) for representing video events with the motivation that the individual component densities of GMM could model some underlying hidden visual attributes and propose a ℓ2-regularized logistic Gaussian mixture regression approach, which is also called LLGMM classifier, for a more generic and complicated MED. We also propose an efficient iterative algorithm, which uses gradient descent, a standard convex optimization method, to solve the objective function of LLGMM. Finally, extensive experiments are conducted on the challenging TRECVID MED 2012 development dataset. The results demonstrate the effectiveness of the proposed LLGMM classifier for MED.
AB - Multimedia event detection (MED) is one of the most important branches of multimedia content analysis. Current research work on MED focuses mainly on detecting specific events, such as sport events, news events and suspicious events, which is far from achieving a complicated and generic MED due to the fact that these events usually contain a lot of visual attributes, such as objects, scenes and human actions. Being different from visual features, visual attributes are hidden classes to event detectors and event classifiers. Hence, proper representation of these visual attributes could be helpful in building a sophisticated and generic MED. In this paper, we use Gaussian mixture model (GMM) for representing video events with the motivation that the individual component densities of GMM could model some underlying hidden visual attributes and propose a ℓ2-regularized logistic Gaussian mixture regression approach, which is also called LLGMM classifier, for a more generic and complicated MED. We also propose an efficient iterative algorithm, which uses gradient descent, a standard convex optimization method, to solve the objective function of LLGMM. Finally, extensive experiments are conducted on the challenging TRECVID MED 2012 development dataset. The results demonstrate the effectiveness of the proposed LLGMM classifier for MED.
KW - Gaussian mixture model
KW - LLGMM classifier
KW - Logistic regression
KW - Multimedia event detection
KW - ℓ Regularization
UR - http://www.scopus.com/inward/record.url?scp=84939571426&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84939571426&partnerID=8YFLogxK
U2 - 10.1007/s00521-014-1810-y
DO - 10.1007/s00521-014-1810-y
M3 - Article
AN - SCOPUS:84939571426
VL - 26
SP - 1561
EP - 1574
JO - Neural Computing and Applications
JF - Neural Computing and Applications
SN - 0941-0643
IS - 7
ER -