A model-based approach for geometrical correction of optical satellite images

摘要

We present an algorithm that corrects optical satellite data by modelling the geometry of image acquisition. The model includes a Keplerian orbital model for the satellite motion, and estimates of the six orbital elements are given as input to the model from ephemeris data. A number of ground control points (GCPs) are then introduced to improve the estimates and correct for attitude angle variations. While similar approaches assume that just one line is acquired at a time, our algorithm works for the more general case where several parallel, rigid lines are acquired at the same time. In our approach the GCPs are detected automatically in the image to be corrected. The performance of the correction procedure was investigated on three kinds of data: NOAA AVHRR images, system-corrected Landsat TM image, and an image with pixel size similar to Earth Observation System (EOS) Moderate Resolution Imaging Spectro-radiometer (MODIS) derived from the system-corrected Landsat TM image by re-sampling. The error is estimated by using a cross-validation approach. For the NOAA AVHRR images and the image similar to EOS MODIS, we obtained satisfactory results with an rms error of about one pixel unit. The model has to be further improved for high-resolution data, such as Landsat TM.