The automated extraction of environmentally relevant features from digitalimagery using Bayesian multi-resolution analysis

In this paper, we discuss the use of hierarchical tree-structured Bayesian networks for integrating knowledge concerning contextual relationships betw een environmentally relevant features extracted from digital imagery at mul tiple resolution scales. In our model, conditional probability distribution s over continuous valued observations are parameterized using a mixture of multivariate Gaussian distributions. Separate classifiers for pixels and gr oups of pixels are used as sub-components of the overall model. The Bayesia n formalism allows models to be composed in a systematic and statistically sound manner. We illustrate how this approach can be used to resolve ambigu ity leading to classification errors and thus improve techniques for the cl assification of land use from aerial imagery. We present an example relevan t to ecosystem analysis, the monitoring of urban growth and the automatic g eneration of input parameters for hydrologic models. (C) 2001 Elsevier Scie nce Ltd. All rights reserved.