Polarimetric SAR image classification-exploiting optimal variables derived from multiple-image datasets

Polarimetric SAR image classification remains an important research area. Various methods continue to be developed for specific applications. Instead of focusing upon one specific problem, we have developed nonlinear dimensionality reduction techniques that extract information from inherently high dimensional datasets. We present computationally tractable, suboptimal extensions of the full nonlinear dimensionality reduction techniques. We apply these suboptimal techniques to polarimetric SAR image classification. Comparisons will be made between optimal and suboptimal techniques, and, as a reference, to standard statistical Wishart classifiers.