Have you ever wanted to find an image or sound to illustrate an abstract concept? Or an image that is metaphorically associated with some text rather than described by the text? This talk will present research on how one can retrieve (text-annotated) images that are related in various ways to a text query, and organize them in a user interface. The method identifies various possible 'interpretations' of the input query using word sense disambiguation techniques (i.e., clustering), and generates an expanded query for each interpretation. The research is an application of lexical association scores (including co-occurrence and similarity scores). This kind of search could eventually have applications in better image search engines, or in the automatic illustration of news articles.

Watch the webcast replay >>