Classification of Non-edited Broadcast Video Using Holistic Low-level Features (2002)
| Venue: | Proceedings of IWDC'2002 |
| Citations: | 2 - 1 self |
BibTeX
@INPROCEEDINGS{Roach02classificationof,
author = {Matthew Roach and Li-qun Xu and John Mason and Martlesham Heath and Ipswich Ip Re},
title = {Classification of Non-edited Broadcast Video Using Holistic Low-level Features},
booktitle = {Proceedings of IWDC'2002},
year = {2002}
}
Bookmark
 |
 |
 |
 |
 |
 |
OpenURL
Abstract
A data driven learning based approach has been adopted to classify a video programme into a set of five pre-defined genre including a7orts, cartoons, news, commercials and music. The video database studied is a collection of five-hour non-edited TV broadcast programme of dozens of video clips: one hour long in total for each genre with an even split of training and testing data. This paper reports the classification result for the two media modes - acoustic and visual - used separately, and for a linear fusion of the two modes at the score level. The correaIonding classification accuracy achieved is approximately 74%, 73%, and 87%, reaIectively, given a 30-second decision window, or for every 30 second non-overlap video segment an answer of its identity is provided Further analysis of the confusion matrices of the classification errors is given. The extension of the approach to the open-set video genre verification and semantic scene segmentation and classification is expected.
Citations
| 117 | D.: Content-based representation and retrieval of visual media: A state-of-the-art review - Zhang, Petkovic - 1996 |
| 55 | A survey on the automatic indexing of video data - Brunelli, Mich, et al. - 1999 |
| 46 | Automatic recognition of film genres - Fischer, Lienhart, et al. - 1995 |
