Mining for image content

Abstract

This paper provides an overview of our eorts to develop algorithms and systems that are able to \mine" useful information from large image collections. One of the core capabilities targeted is the ability to nd instances of specic objects in images. Such a capability would greatly enhance analysis of NASA's existing image archives and, if coupled with an autonomous agent that is able to act on the mined information, would enable new forms of scientic exploration. Keywords: image mining, object recognition, content-based retrieval, distributed systems, database, Diamond Eye, space exploration, queries, knowledge discovery, digital libraries. 1 Introduction There are two trends developing within JPL and NASA that have created a demand for systems that can automatically analyze images and extract semantically meaningful information. First, the amount of data returned from spacecraft missions has greatly increased due to improvements in image acquisition, communications, and storage...