Automatic Content Indexing of Image Data Bases

This paper examines current research on the automatic content indexing of image data bases. Digital images are becoming increasingly prevalent as computational technology becomes commonplace. The volume of the image information being produced along with its divergence in form from prior information retrieval work has lead to a number of innovative approaches. Most of these systems do not take full advantage of prior work in automatic indexing of text. This paper reviews the automatic image indexing work in the light of developments in automatic text indexing. It reviews the tasks that are performed in an information retrieval system. These include: identification of content, conceptual structure and query formation. The parallels between lexical analysis and image analysis are explored. Term selection and organization as well as document classification in text systems are reviewed and the concepts are the relevance to image indexing is discussed. Representative automatic image indexing system discussed in the paper are IMAID, MMIS-2, Aerial Image interpretation and MULTOS.\ud It is that the quality of image indexing could be improved by introducing IR metrics for the automatic selection of index features. The use of a thesaurus, as is common in text IR systems, is virtually unknown in image IR but could be readily exploited. Likewise the potential of automatic term and document classification has not been explored. There is an opportunity for research into the seamless integration of text and image indexing and retrieval