Fusion model for ontology based image retrieval

Abstract

Data fusion is the process of combining multiple sources of information to produce better results compared to using the source individually. This paper applies the idea of data fusion to semantic image retrieval, which combines the ranking scores between ontology based and keyword based semantic image retrieval model. Although the evaluation shows that the overall performance of the ontology based model is higher than that of the keyword based model, the results analysis reveals that the performance of ontology based model is in direct relation with the implicit information relies within the query and annotation text. If the annotation contains less meaningful information, the ontology based method performs very poorly, thus affecting the relevancy of semantic chromosomes. This further affects the performance of the similarity measure and the quality of the retrieval results. As a result, user queries return fewer results than expected, as they get much lower similarity value than they should. Whereas, keyword-based search would perform better in these situations. To deal with this drawback, this paper proposes to combine the results coming from the proposed ontology-based retrieval model and the result returned by traditional keyword-based model. The combined model is evaluated using both traditional IR measures.