Capturing Diverse Usage Contexts forWeb-Based Images

Abstract

The rapidly increasing number of image collections on the Internet require improvements in our image retrieval systems if these collections are to be easily available for browsing and searching by information seekers. Current techniques for image retrieval have known shortcomings that make it difficult to search for images based on their semantic content. We propose that an increased use of image context information can improve identification of image semantics, and may thus contribute to closing the gap between user needs for semantic image retrieval and the capabilities of current image retrieval systems. In this paper we present a new category of image context, called usage context, describe how usage context from multiple sources can be combined, and show how usage context information can be specified using a new context descriptor vocabulary CTXT.