抄録
In the conventional scheme of geometric correction, the selection of ground control points (GCPs) is a major obstacle to automation of geometric correction. The author proposed to use the correlation between ortho-rectified satellite image and simulated shading relief image made from DEM as the evaluation function for optimizing the geometric transformation. Although we obtained accurate and robust parameters of transformation of relatively narrow area (18km by 18km), rough guess of the parameters was needed to initiate the successful optimization process. In this paper, we consider a full automation method for precise geometric correction. It is noted that land pixels are easily discriminated from water pixels in the satellite images of infrared bands, which are also identified on the digital map. The classification error of land/sea discrimination is used as the evaluation function for the optimization. Stratified random sampling of test points make the rough guess of the parameter unnecessary.
