Identifying human settlements in remotely-sensed data: The automated production of maps of human settlement from recent satellite images is essential to studies of urbanization and population movement. The spectral and spatial resolution of such imagery is high enough for accurate identification of human settlements, but we need to process vast amounts of data quickly. We proposed using a multi-level approach that started with simple, but fast, techniques at the lowest level, and progressively moved to more complex approaches as we reduced the size of the data that required processing. We demonstrated our approach using IKONOS 4-band and panchromatic images.
SBOR – Similarity-based object retrieval: Computer simulations generate vast quantities of spatio-temporal output that can be challenging to explore using visualization alone. A display of an interesting region in one simulation may well prompt a scientist to ask if there were other similar regions in the same or in other simulations. We borrowed ideas from the field of content-based image retrieval (also called query by image content) to demonstrate that we could indeed address such questions, provided we carefully extracted the features representing the region of interest.
Validation of computer simulations: Validation is the process of checking how close a computer simulation is to reality, for example by comparing the simulation with experiments. Since the simulation output could be a two-dimensional unstructured grid, while the experimental data may be in the form of images, a direct comparison is not an option. Using the Richtmeyer-Meshkov instability as an example, we showed how we could extract features from both the simulation grid data and the noisy experimental images to validate the simulation.