The availability and demand for images and other spatially-referenced data is increasing dramatically. This presents huge, interdisciplinary challenges for data retrieval, analysis and interpretation. Effective visual information management requires symbolic abstraction at varying spatial scales and at varying levels of semantic detail. This project focuses on two key issues. First, abstraction is used to dynamically match symbolic representations of both data and knowledge to the computational requirements of the task at hand. Second, abstraction is used to establish context for fine-level, content-based analysis of the underlying raw visual data. The testbed application is geographic information systems (GIS) and remote sensing with a specific focus on forest resource management. In the long-term, the technology developed will impact a wide range of application areas including commerce, education, medicine and scientific visualization.
The project integrates research in computational vision, databases, computational geometry and knowledge-based systems. It relates directly to two of the three broad topics identified in the Intelligent Computation theme, namely ``knowledge representation and reasoning techniques'' and ``efficient implementation techniques for storing and processing large knowledge bases, spatial and visual databases...'' It advances applications of intelligent data analysis and management to domains where both the data, in the form of images, and the associated knowledge bases are largely unstructured.