MOSAIC: A fast multi-feature image retrieval system

Content-based image retrieval plays an important role in many multimedia ap plications. Images are typically retrieved based on a given sample image, a sketch or a simple description of the content. The rate of this retrieval is undeniably gaining importance as databases increase constantly in size. In this paper, we present MOSAIC, an image retrieval system that we have de veloped. In MOSAIC, an image is represented by a set of clusters, each of w hich captures information on multiple features - the "color" of the cluster , the "size" of the cluster and the "spatial" location of the cluster. We a lso propose an index structure, to facilitate speedy retrieval of images. T he structure is multi-tier with each tier dealing with one feature. In this way, images that are dissimilar in the higher tier can be pruned away imme diately. We implemented and evaluated MOSAIC, and our results show that the system can retrieve relevant images effectively and efficiently. (C) 2000 Elsevier Science B.V. All rights reserved.