LANDSAT images are most useful for geological lineament extraction, especially for large-scale lineaments. However, because of the nature of the LANDSAT orbit, there are almost no lineaments in the same direction as the sun angle. In this report, a new technique which can generate virtual LANDSAT images with any sun directions is described.
In order to elminate shadows from the original LANDSAT image, a shadow-free land cover classification was first applied. Shadow areas were classified into four categories using spectral features. Pixel values in each categories of the original image were scaled to the values of corresponding categories in no-shadow areas, by coinciding the mean and standard deviation of each spectral bands.
Next, a three-dimensinal model of the object area was constructed from the National Land Information Data Base. Assuming a specific sun elevation and azimuth, the shadowed areas can be calculated from this model. A cosine correction was made using the angle between the sun light axis and the plane direction. In addition, reflective were calculated according to a three dimentionl reflective intensities model. An artificially shadowed LANDSAT image with specific sun angle was thus obtaind.
For lineament extraction, a sun azimuth was selected perpendiculer to the real sun azimuth. A complete set of lineaments was obtained by combining the two results extracted from these two images.
In conclusion, a technique to generate a virtual LANDSAT image with illumination at any sun angle was established. With the aid of this virtual image, lineament extraction becomes more complete.
View full abstract