A statistical approach for topographic correction of satellite images by using spatial context information

The geometric and systematic errors associated with the acquisition and coregistration of a satellite image and digital terrain model (DTM) will significantly affect the results of topographic corrections. The conventional pixel-based topographic correction methods have not handled these errors well. The corrected images, although exhibiting no significant residual topography in average, usually show high and nonhomogenous variability across the scene. The authors propose a contextual approach for minimizing this artifactual and undesirable feature in the corrected images. The new approach compensates the topographic shading and shadowing by using the local reflectance estimated from the spatial contexts. Since the noises have much less effects on the extracted contextual information, errors are reduced for the estimated reflectance and the signal-to-noise ratios are improved on the shaded slopes. As a result, not only is the variance lower and spatially more homogenous, the fine textures and community boundaries are also well preserved on the corrected images. For the purpose of image interpretation, the reduced variability for each cover type may lead to significant improvements of land cover differentiation. In the testing site of a forested scene, for example, the overall classification accuracy has improved about 9% in the contextually corrected image over the conventionally corrected images