In previous image registration research, edge image correlation techniques were applied to Landsat-1 multispectral digital image data. Translations of edge images of an earlier area of interest were performed over a search area edge image at a later time to determine the location of the area of interest in the search area. Plots of the correlation surface indicated sharp peaks for successful searches and an accuracy of ±1 pixel or line. The following three new requirements were suggested: (1) because clouds or their shadows are time-dependent image features, removal of cloud or cloud shadow information from the edge images before image correlation takes place; (2) accurate representation of geometric shapes of image features; (3) fractional pixel registration accuracy over subregions of the area of interest. Solutions to these problems were obtained. Distinguishing features of the geometric correction solution are the automatic location of the ground control points needed to determine correction parameters and the use of the geometric centers of edge image chips as the ground control points.
Date of this Version