Extracting contextual information from multiuser systems for improving annotation-based retrieval of image data
Johanna Vompras, Thomas Scholz, Stefan Conrad · 2008
In this paper, we present an approach for incorporating contextual knowledge into a multiuser image retrieval system which is based on annotations. Although the most existing keyword-based systems are expanded by conceptual knowledge (e.g. ontologies) modeling the topics in which the user is interested in, there still remain some unresolved problems, like existing differences in interpretation of image contents or inconsistencies in keyword assignments among different users. In our approach, multiple sources of information which are modeled as different annotation ontologies are brought together in order to extract contextual information, and thus attenuate users' subjectivity in content description. Finally, we evaluate our introduced approach on a real data set of sports images. The experiments show that our approach provides considerable retrieval quality, already in the first search iteration, which makes an additional query refinement dispensable. The results can even be further improved by applying lexical analysis for strings and error elimination methods.