Using a game to evaluate image retrieval, organization, and annotation

Methods of retrieving images that incorporate human- generated metadata, such as keyword annotation and collaborative filtering, are less vulnerable to the semantic gap than content-based image retrieval. However, generating such metadata is time-consuming, expensive, and difficult to evaluate. This paper discusses an interface for image retrieval that is able to simultaneously compose queries based on content- based features, keywords, and collaborative relations. We propose using a game metaphor to evaluate the system.