Structured High-Level Indexing of Visual Data Content

Unstructured manual high-level indexing is too open-ended to be useful. Domain-based classification schemes reduce the variability of index captions and increase the efficiency of manual indexing by limiting the indexer’s options. In this paper, we incorporate classification hierarchies into an indexing superstructure of metadata, context and content, incorporating high-level content descriptions based on the ternary fact model. An extended illustration is given to show how metadata can be automatically extracted and can subsequently help to further limit the indexer’s options for context and content. Thus, this structure facilitates the indexing of high-level contents and allows semantically rich concepts to be efficiently incorporated. We also propose a form of data mining on this index to determine rules that can be used to semi-automatically (re)classify images.