Automatic generation of textual image collection descriptions

  • Authors:

  • Randi Karlsen;David Sundby;Joan Nordbotten

  • Affiliations:

  • University of Tromsø, Tromsø, Norway;University of Tromsø, Tromsø, Norway;University of Bergen, Bergen, Norway

  • Venue:
  • Proceedings of the 3rd International Conference on Web Intelligence, Mining and Semantics
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

Digital photo collections are growing rapidly in number and size, making location of relevant images increasingly difficult. Since very few image collections have a description of their content, an image seeker must manually locate image collections and use that collection's image retrieval system to determine if there are any relevant images for his/her need. Manually describing a large and growing image collection that may contain a variety of diverse themes, is a near impossible task. Thus some techniques for automatic generation of image collection descriptions are needed. Further, since most image retrieval systems use text-based queries, it is important that the collection description also be text-based. In the following, we present a system that can automatically generate an image collection description based on available metadata from all images in the collection. These metadata are extended, using information and services available on the Internet, to provide a rich description of the collection as a whole.