Relevance Feedback and Term Weighting Schemes for Content-Based Image Retrieval

This paper describes the application of techniques derived from text retrieval research to the content-based querying of image databases. Specifically, the use of inverted files, frequency-based weights and relevance feedback is investigated. The use of inverted files allows very large numbers (≥ O(104)) of possible features to be used, since search is limited to the subspace spanned by the features present in the query image(s). Several weighting schemes used in text retrieval are employed, yielding varying results. We suggest possible modifications for their use with image databases. The use of relevance feedback was shown to improve the query results significantly, as measured by precision and recall, for all users.