Feature normalization and likelihood-based similarity measures for image retrieval

Distance measures like the Euclidean distance are used to measure similarit y between images in content-based image retrieval. Such geometric measures implicitly assign more weighting to features with large ranges than those w ith small ranges. This paper discusses the effects of five feature normaliz ation methods on retrieval performance. We also describe two likelihood rat io-based similarity measures that perform significantly better than the com monly used geometric approaches like the L-p metrics. (C) 2001 Elsevier Sci ence B.V. All rights reserved.