Image indexing in article component databases

It is often necessary to compare data-rich charts, tables, diagrams, or drawings rather than the articles that contextualize that data. The objective of this research has been to create a database of non-textual components (here, maps) that are searchable independently of the articles from which they are taken, with the option to view the source articles. The method mines words from the articles that are near or associated with each component map, and these mined words become the basis of region, time, and subject indexing. The evaluation showed that automatic indexing of the component maps by these three facets works well, and indicates that a large-scale component database following this model is viable.