L-GEM based RBFNN for news anchorperson detection with Dominant Color Descriptor

Author

Xiao, Wei ; Ng, Wing W Y ; Chan, Patrick P K ; Yeung, Daniel S.

Author_Institution

Machine Learning & Cybern. Res. Center, South China Univ. of Technol., Guangzhou, China

Volume

2

fYear

2010

fDate

11-14 July 2010

Firstpage

763

Lastpage

768

Abstract

News reports in TV provide important and timely information about the city and the world to citizens. Moreover, data mining and indexing of news video clips provide a good source of information. However, news video usually consists of more than one news story. One must split them into individual news before indexing. Owing to the nature of news reports, the news anchorperson usually appears in the transition of two news story. Therefore, we propose a new method to find the news anchorperson shot in news video. The MPEG-7 Dominant Color Descriptor (DCD) is adopted to describe video frames. Radial Basis Function Neural Network (RBFNN) trained by minimizing the Localized Generalization Error (L-GEM) is adopted to classify the occurrence of news anchorperson in video frames. Experimental results show that the proposed method is accurate for different news videos from different TV stations.

Keywords

image classification; image colour analysis; radial basis function networks; video signal processing; DCD; L-GEM; MPEG-7 dominant color descriptor; RBFNN; data mining; dominant color descriptor; localized generalization error; news anchorperson detection; radial basis function neural network; Biomedical monitoring; Biosensors; Image color analysis; TV; Temperature measurement; Temperature sensors; Training; Anchorperson; L-GEM; News video; RBFNN;

fLanguage

English

Publisher

ieee

Conference_Titel

Machine Learning and Cybernetics (ICMLC), 2010 International Conference on

Conference_Location

Qingdao

Print_ISBN

978-1-4244-6526-2

Type

conf

DOI

10.1109/ICMLC.2010.5580574

Filename

5580574