Structuring the visual content of digital libraries using CBIR systems

We describe the benefits of CBIR (content-based image retrieval) systems for retrieval and organization of visual content in databases of digital libraries. Unlike text-based retrieval systems which work with manually annotated keywords as metadata, CBIR systems use automatically extracted numerical representations of perceptual features such as color, texture or shape. This technique enables users of digital libraries to retrieve visual content without the help of textual metadata, which in some cases may be either nonexistent or not sufficient for a special purpose. It is our approach to support maintainers of digital libraries in organizing large image volumes using CBIR systems. Finding clusters of similar content or providing clusters with suitable keywords are new application contexts of CBIR. We propose a method for structuring visual content and evaluate retrieval results of CBIR systems with regard to their applicability to this method.