Algorithms for the Automated Matching of Remotely Sensed Imagery

Remote sensing image data from satellites can be a valuable tool used in the assessment of changes in the environment. These data can be used to characterize land usage, to determine change in land usage, or to plan for future land use. Before these image data can be used in assessing change, however, they must often be registered to a common geometric base. The geometric registration of remotely sensed imagery requires many steps. Until recently, much of this registration work at the United States Geological Survey\u27s Earth Resources Observations Systems (EROS) Data Center was performed manually, a time consuming process. Recent projects, however, demanded that a faster method be developed. This study describes the methods of cross correlation, normalized cross correlation, binary edge cross correlation, filtered phase correlation, and methods to edit matched points as part of an automated geometric registration process. When applied to Landsat Multi-Spectral Scanner data in a hierarchical matching process, accuracies of less than one pixel can be achieved. When applied to Synthetic Aperture Radar imagery, 3/4 pixel accuracy can be achieved. This study integrated these techniques into the Land Analysis System, the production image processing system used at the EROS Data Center