Enhanced image annotations based on spatial information extraction and ontologies

Current research on image annotation often represents images in terms of labelled regions or objects, but pays little attention to the spatial positions or relationships between those regions or objects. To be effective, general purpose image retrieval systems require images with comprehensive annotations describing fully the content of the image. Much research is being done on automatic image annotation schemes but few authors address the issue of spatial annotations directly. This paper begins with a brief analysis of real picture queries to librarians showing how spatial terms are used to formulate queries. The paper is then concerned with the development of an enhanced automatic image annotation system, which extracts spatial information about objects in the image. The approach uses region boundaries and region labels to generate annotations describing absolute object positions and also relative positions between pairs of objects. A domain ontology and spatial information ontology are also used to extract more complex information about the relative closeness of objects to the viewer.