Automated generation of a digital elevation model over steep Terrain in Antarctica from high-resolution satellite imagery

The automated generation of a digital elevation model over the Antarctic using stereo matching high-resolution satellite images is a challenging task. Moreover, the homogeneous radiometry in the icy environment and the strong geometric dissimilarity between stereo pairs over the steep terrain limit the use of area-based matching techniques. To overcome this issue, we propose template matching with image transformation in order to reduce the geometric dissimilarity. First, we generated epipolar resampled images to ensure the ease of estimation and handling of image dissimilarities from various viewing directions. We then utilized the normalized cross-correlation (NCC) and transformed the image patches within the matching window along the sample, line, and diagonal directions in order to improve the match rates within the steep areas. Furthermore, we tested the proposed method using Antarctic IKONOS stereo images and found that the overall matching success rate improved from 93.5% to 97.0% for all image pixels. We then computed the success rates over an area in which the NCC produced a low elevation point density and observed a more significant improvement from 58.7% to 79.26%. When compared to the manually generated elevation, the maximum vertical difference improved from 11.4 to 4.7 m. With these improvements, we can build a 1-m resolution elevation model over the glaciated high relief terrain.