Georegistration

From MilcordWiki

Overview

AIR (Automated Image Registration) is our georegistration solution based on an object configuration similarity approach. Our solution - a departure from traditional approaches – uses abstract spatial relations (e.g. three square buildings forming an orthogonal triangle with a river running between them) as matching features, and transforms the registration problem into a spatial similarity assessment problem. Our approach results in unparalleled pull-in range as coarse location data are adequate to support the recovery of registration information for a configuration.

Need

Conflation and co-registration are critical applications for co-registration of datasets differing in scale, resolution, sensor spectrum range (e.g. optical with IR or thermal imagery). Traditional georegistration solutions in the geospatial, photogrammetric, and computer vision communities select a set of control features in an existing database and identify the same features in an incoming dataset. However, these registration algorithms are not robust due to variations in scale, resolution, and sensor characteristics, where the identification of control features becomes a complex and challenging process.

Approach

Our scene similarity function combines different similarity metrics for individual object shapes and relations between them like orientation, and distance. We use a position matrix that captures the relative "left-of" or "right-of" the features for the orientation similarity metric to compute the normalized correlation coefficient for each query/image scene. For distance similarity, we employ a distance ratio matrix that captures the ratios of the distances between pairs of objects. We employ configuration agents that find the optimal configuration containing a target, and similarity agents that select the best set of similarity metrics for given object configuration measurements.

Benefits

Unlike automated feature based registration schemes that pick a few control features to co-register two datasets, AIR does not need precise individual features as a set of coarse features is sufficient for georegistration in AIR.

Applications

NASA applications include image processing applications in Lunar, Mars and other planetary missions, Earth Observing System Data and Information System (EOSDIS) Core System (ECS) Synergy Program, NASA Distributed Active Archive Centers (DAAC), Remote Data Store (RDS). DoD applications include Distributed Common Ground System and similar programs focused on image exploitation. Commercial applications include imaging data publishers and their network of partners serving the remote sensing and satellite imagining market.

References

• Stefanidis, A., Agouris, P., Bertolotto, M., Carswell, J., Georgiadis, C.: Scale and Orientation-Invariant Scene Similarity Metrics for Image Queries, International Journal of Geographical Information Science , Vol. 16, No. 8, pp. 749-772, December 2002.
• Windholz, T.K., Beard, M. K., and Goodchild, M.F. (2000), Data Quality: A Model for Resolvable Objects. In: Advances in Spatial Data Quality, Shi, W., Goodchild, M.F., and Fisher, P.F., eds., Taylor & Francis.