Optimal polarimetric decomposition variables-non-linear dimensionality reduction

Polarimetric SAR image analysis often depends upon proper identification of the relevant degrees of freedom for the problem at hand. Employing physical models of particular scattering processes simplifies identification of the appropriate polarimetric variables. Determining how well variables chosen on the basis of a particular model describe the region of applicability of that model is difficult. Here we attempt in a model independent manner to identify "optimal" variables to both segment an image and highlight the variation within each segment. The method presently employed is non-linear dimensionality reduction.