Unifying Different Users´ Interpretations and Levels of Abstraction for Improving Annotation-based Image Retrieval

The proceeding application of multimedia information systems has brought the need for developing efficient querying and browsing methods for large image repositories. One solution to overcome the semantic gap between low-level visual features of images and high-level human perception is to generate semantic annotations for images to describe their contents. Therefore, the broad variety and the lack of standards among different annotation tools make it necessary to develop an annotation model supporting the unification and integration of different annotations created by users having different background knowledge. In this paper we present a multi-level annotation model which considers the several levels of abstractions at which content descriptions are assigned and we show how it can be utilized in sophisticated retrieval systems to support image retrieval at the semantic level.