Feature-based indexing is one method that can be used to provide the needed access to first-hand, unbiased primary subject mat-ter in visual information systems. The diversity and dynamism of user information needs and the inherent characteristics of visual information however make the process of indexing quite challenging and thus some unique approaches are required. From an analysis of the special nature of visual information, and of user information needs, the paper identifies the requirements for providing acceptable levels of performance in visual information retrieval. To meet the identified requirements, the paper pro-poses a uniform indexing framework, which is hierarchical, general and extensible, supporting both multiple indices and multilevel indices, while providing the needed constructs for similarity and relevance grouping. Candidate features that meet the requirements of robustness and uni-form indexing at the various levels of the index hierarchy as may be needed in the uniform framework are also indicated.
展开▼