Iterative spatial filtering for reducing intra-class spectral variability and noise

Intra-class variability and noise are obstacles that obscure subtle differences between spectral classes in hyperspectral imagery. This paper presents an iterative adaptive smoothing filter (IAS), which considers inherent spatial characteristics of image classes and the assumed random nature of pixel to pixel noise to minimize intra-class variability and noise. IAS makes use of standard hyperspectral spectral similarity measures, spectral angle and root-mean-squared error, to calculate and apply weighting functions to filter image pixels. Using a small window assures that spatially independent classes with subtle spectral differences can still be distinguished. The result is a change in the internal density distribution of the data volume (intra-class variability and noise), but the overall volume undergoes little change (inter-class variability). The usefulness of the filter is illustrated with simulated and real hyperspectral data.