2004 Volume 24 Issue 4 Pages 378-386
For geometric correction of satellite images over rugged terrain, we have proposed an optimaization method in which a simulated irradiance image from digital elevation model is used to determine parameters of geometric transform equation. In this paper, we propose another optimization method which supplements the former method as it is used for coastal area.
It is noted that difference between land surface and water surface is the most distinctive feature in satellite images, particularly, in middle infrared bands. Instead of careful selection of GCPs in conventional geometric correction methods, a number of evaluation points are selected both from sea and land over a digital map by using stratified sampling method with regard to distance from coastline. Then, these samples are checked at the corresponding image coordinates calculated by the geometric transform equation. By using misclassification rate as an evaluation function, the parameters (scene center displacements) in the transform equation as well as an threshold between sea and land are optimized.
We apply the method to 13 system-corrected Landsat images to check their accuracy and computational efficiency. It is shown that this method can provide scene center displacements with the accuracy of few pixels and with very high computation efficiency. The accuracy could be improved by selecting sample points more carefully.
