Context-based image retrieval: A case study in background image access for Multimedia presentations
Abstract
Conventional approaches of image indexing and retrieval from digital libraries include content-based, metadata-based, and keyword-based approaches. This paper addresses a different way of image retrieval motivated by real-life applications for an intelligent system that can automatically select appropriate background images from textual passages. We explored techniques for developing automatic image-retrieval systems based on essential contextual information of a textual passage. We propose a framework that applies semantic role labeling techniques and a commonsense knowledge base, ConceptNet. The primitive results indicate that the proposed methodology has a potential on applications with textual passages that describe things and events that are regularly seen in every day life. However, for fantasy tales that describe truly fictitious things and events, the use of ConceptNet does not allow to obtain accurate results.
Origin : Files produced by the author(s)
Loading...