To use aerial imagery in the rendering of terrain as it would appear at different times of day, we need to minimize the luminance variability in the source imagery that is due to illumination effects at the time the imagery was acquired. If we had a way to recover surface albedo from luminance, this would also aid in determining what sort of surface cover was present at a given location in the image. Given Lambertian surfaces, a known distribution of illumination, and surface orientation at every point, it is straightforward to determine surface reflectances. In practice, we know none of these properties. Surface reflectance is far from Lambertian, illumination depends not only on sun angle but also on complex, weather dependent variations in sky brightness, and DEMs provide only low resolution information about surface orientation.
Nevertheless, shading effects can be reduced by applying a normalization based on the cosine of the angle between the approximate surface orientation, as specified in a DEM, and an estimate of the direct illumination direction. Sun angle is often provided with satellite data. For USGS orthoimages, it must be estimated from the imagery. Computer vision shape-from-shading methods can be used to solve this problem [3]. If shadows are present and one or more matches can be established between points on shadow generating contours and the corresponding point on the trailing edge of the shadow, then the direction of direct illumination can be inferred from the DEM-specified elevations of the corresponding points.