We develop a fast and accurate method that is able to automatically select and match a large amount of ground control regions (GCRs) for orthorectifying remote sensing imagery. This new method is comprised of four modules, namely automatic extraction of GCRs, fast image-to-image matching, iterating and filtering of GCRs, and rigorous orthorectification. We assess the accuracy of this new method by processing the high-temporal- and high-spatial-resolution Formosat-2 imagery. Results show that the accurate orthoimage with a root mean square error of less than 1.5 pixels can be automatically generated from one standard Formosat-2 image (covering 12 km x 12 km) in 55 minutes. This new method has been incorporated into the Formosat-2 automatic image processing system and has been used to produce orthoimages on a daily-basis.