NONPARAMETRIC MODELING OF RADIANCE IN HILL COUNTRY FOR DIGITAL CLASSIFICATION OF AERIAL PHOTOGRAPHS

This article presents a method to account for topographic, lens falloff, and hot-spot effects when digitally classifying an aerial photograph. Nonparametric models of radiance have been used in conjuction with a Skidmore-Turner classifier to improve classification accuracy. Without the nonparametric models, the Skidmore-Turner classifier achieves an overall 76% classification accuracy of a test image. This is increased to 91% when nonparametric models, using independent variables, view vector, and cosine of incidence angle, are combined with the Skidmore-Turner classifier. Although there is a significant improvement in accuracy, the predominant improvement seems to come from the nonparametric models effectively giving spatially local a priori probabilities rather than from a purely geometric modeling of radiance.