Indexing of On-line Multimedia Repositories using Semantic Expansion and Visual Analysis

Dr. Tomas Piatrik, Queen Mary University of London

Date: 16/03/2012
Time: 15:00-15:45
Room: H-F 114

Manually annotating large scale content such as Internet videos is an expensive and consuming process. Furthermore, community-provided tags lack consistency and present numerous irregularities. To simplify and improve the process of tagging online videos, which are unbounded to any particular domain, a framework for predicting general tags from the associated textual metadata and visual features will be presented.

Moreover, in order to test the proposed framework in a more specific scenario, it is applied in a geo-tagging task, which aims to assign geographic coordinates to videos using the same features. The proposed approach first extracts named entities exploiting complementary textual resources such as Wikipedia and Wordnet. Extending the functionalities of the in-built GATE named entities, the framework integrates a bag-of-articles algorithm for effectively extracting relevant articles from the Wikipedia pages.

Two experiments will be presented for validation of the framework. The first experiment evaluates the framework against MediaEval 2010 Wild Wild Web dataset for the tagging task. The second experiment use MediaEval 2011 Placing Task dataset for the geo-tagging. Both datasets consists of large collections of Flickr videos.