Metadata Information Generation.

Peter Bajcsy

Proceedings of NIMA/NSF Motion Imagery Workshop, Herndon, VA, November 29-30, 2001.

The proposed goal of this project is to generate metadata information from motion imagery and other supplemental data for content retrieval. In this effort, we address most of the following issues that we believe will be critical in generating metadata from motion imagery in the near future. The critical issues include (1) developing a unified software framework, (2) establishing computational benchmarks for metadata generation, (3) exploring methods for computer learning from annotated imagery, (4) building probabilistic hierarchical organization of metadata, (5) engaging tools for processing heterogeneous multi-modal data sets, (6) generating metadata from compressed imagery and (7) defining a standard format for metadata generation.

A proposed project of Seismic Hazard Mapping illustrates how some of the aforementioned issues could be resolved.