|
Compared to spaceborne remote sensing images, the airborne multispectral scanning images have the advantages of higher spatial and temporal resolutions. However, more severe geometric distortions often present in the airborne images owing to their low flight altitudes. Rectification of remote sensing images is often implemented using the polynomial trend mapping (PTM) model which employs a set of ground-control-points (GCPs) for estimation of model coefficients. The total number and spatial distribution of the ground control points play a decisive role in the accuracy of image rectification. Among many sources of image distortion, the relief displacement is of most significant, particularly in areas of rugged terrain; therefore, the objective of this study is to establish a GCP selection scheme that considers the spatial variation characteristics of the terrain elevation. Our approach of GCP selection is based on a premise that a GCP set that yields high interpolation accuracy of terrain elevation will also result in high image rectification accuracy. The premise is particularly applicable for areas of rugged terrain where relief displacement is the dominant type of geometric distortion. We first select a set of uniformly distributed initial GCPs, and subsequent GCPs are determined based on a kringing approach of elevation interpolation using digital terrain model (DTM) data. Both check point validation and cross validation were implemented to assess rectification accuracies. The accuracies of elevation interpolation and image rectification are found to be consistent. A criterion for determining the total number of GCPs needed for image rectification is also suggested.
|