Challenge of Image Retrieval (CIR)
5 February 1998
Tony Stone Images have designed a textual classification structure and an image retrieval system to store and retrieve pictures within the stock photography domain. The image retrieval system has been designed from the stock photography domain-specifics with facilities to search for different objects and their attributes within an image, and on relations between these objects. The retrieval system is text-based with a set of 12, 000 controlled hierarchical keywords in a semantic network structure which has approximately 45,000 additional search terms in the form of synonyms linked to these keywords. The keywords allow for searching both on specific objects within a picture, and for abstract concepts that an image signals as a whole. The keywords are added manually to each image and they are searched on in different fields of the user interface of the retrieval system depending on what type of content is wanted. Furthermore, the image retrieval system deals with ambiguity by prompting the user at the time of searching and it allows for multi-lingual searching in French, German and American English besides British English.