Improving Classification of Video Shots Using Information-Theoretic Co-Clustering

Abstract

Automatic categorization of video shots is very useful in applications of video content analysis and retrieval, such as structure parsing and semantic event recognition. In order to consider the relationships between different video features and provide more accurate similarity measure for video shot classification, in this paper, information-theoretic co-clustering is utilized to group the video shots and features simultaneously. In addition, a Bayesian information criterion is employed to automatically estimate the number of clusters for both the video shots and features. Evaluation on 1374 shots extracted from around 4-hour sports videos shows very encouraging results.