News story clustering using L-GEM based RBFNN

Abstract

With the explosive growth of video resources on the Internet and Internet connected mobile devices, the needs of efficient video retrieval and video clustering are increasing. Video clustering is important because video resources on the Internet may be duplicate and highly similar. To reduce the use of bandwidth, users would like to fetch only one representative video resource instead of a number of highly similar or duplicated videos. In this paper, we first summarize current research development of video retrieval. Then, we propose a new video clustering method based on a Radial Basis Function Neural Network (RBFNN) trained via a minimization of the Localized Generalization Error (L-GEM). The L-GEM provides estimation on the generalization capability of the RBFNN which helps to cluster news video from different channels with higher accuracy. Experimental results show that the proposed method outperforms RBFNN trained without the L-GEM.