An approach for correcting inhomogeneous atmospheric effects in remote sensing images
Two contextual-based approaches for correcting first order atmospheric effects due to atmospheric inhomogeneity and adjacency effect are described. The first method is a modification of the restoration method for the adjacency effect suggested by Richter. The second method is an adaptation of Stenberg's rolling ball algorithm using the mathematical morphology transformation. Evaluation of the proposed method was carried out by noting classification accuracy on the basis that an increase in classification accuracy reflects in improved image quality. Results show that the average accuracy of classification of a scene from Malaysia, with 13 ground truth classes was as follows: (a) uncorrected--86%, (b) adjacency correction--86%, and (c) rolling ball correction--89%. It is clear from the results that the rolling ball method of correction does yield an increase in accuracy although not as substantial as in this case, and there is no reason to doubt that this method could be used to correct a wide variety of images that are affected by inhomogeneity of the atmosphere.