Melanie Aurnhammer , Peter Hanappe , and Luc Steels
We propose an approach that unifies browsing by tags and
visual features for intuitive exploration of image databases. In contrast to
traditional image retrieval approaches, we utilise tags provided by users
on collaborative tagging sites, complemented by simple image analysis
and classification. This allows us to find new relations between data elements.
We introduce the concept of a navigation map, that describes
links between users, tags, and data elements for the example of the collaborative
tagging site Flickr.We show that introducing similarity search
based on image features yields additional links on this map. These theoretical
considerations are supported by examples provided by our system,
using data and tags from real Flickr users.