Automatically organising images using concept hierarchies

In this paper we discuss the use of concept hierarchies, an approach to automatically organize a set of documents based upon a set of concepts derived from the documents themselves for image retrieval. Co-occurrence between terms associated with image captions and a statistical relation called subsumption are used to generate term clusters which are organized hierarchically. Previously, the approach has been studied for document retrieval and results have shown that automatically generating hierarchies can help users with their search task. In this paper we present an implementation of concept hierarchies for image retrieval, together with preliminary ad-hoc evaluation. Although our approach requires more investigation, initial results from a prototype system are promising and would appear to provide a useful summary of the search results.