Report ID
1996-19
Report Authors
Daniel Wu, Divyakant Agrawal, Amr El Abbadi, Ambuj Singh, andTerrence R. Smith
Report Date
Abstract
The management of large image databases poses several interesting andchallenging problems. These problems range from ingesting the data andextracting meta-data to the efficient storage and retrieval of the data. Ofparticular interest are the retrieval methods and user interactions with animage database during browsing. In image databases, the response to a givenquery is not an exact well-defined set, rather, the user poses a query andexpects a set of responses that should contain many possible candidates fromwhich the user chooses the answer set.We first present the browsing model in Alexandria, a digital library for mapsand satellite images. Designed for content-based retrieval, the relevantinformation in an image is encoded in the form of a multi-% dimensional featurevector. Various techniques have been previously proposed for the efficientretrieval of such vectors by reducing the dimensionality of such vectors. Weshow that for even moderately large databases (in fact, only 1856 textureimages), these approaches do not scale well for exact retrieval. However, as abrowsing tool, these dimensionality reduction techniques hold much promise.
Document
1996-19.ps292.06 KB