Ontologies for object-based image retrieval

In this paper, a novel approach to image retrieval is presented. The proposed approach employs a time-efficient and fully unsupervised segmentation algorithm to divide images into regions. Low-level indexing features for each region are subsequently extracted. These arithmetic features are automatically mapped to appropriate qualitative descriptors of the object ontology. The object ontology is also used for describing semantic objects (keywords). By querying for images conforming to the qualitative description of the desired object, clearly irrelevant image regions are rejected; following that, a relevance feedback mechanism, based on neural networks, is invoked to rank the remaining, potentially relevant regions and produce the final query results. The proposed approach bridges the gap between keyword-based approaches, which assume the existence of rich image captions, and query-by-example approaches, which assume that the user queries for images similar to one that already is at his disposal.