Optimal image classification employing "optimal" polarimetric variables

Analysis of polarimetric SAR data often proceeds after first identifying important degrees of freedom in the data, thereby reducing the dimensionality of the problem. Various derived parameters (e.g. entropy, polarization fraction) and/or descriptions of polarimetric data (e.g. scattering matrix, covariance matrix, Stokes Parameters) are employed to effectively reduce the dimensionality of the data while at the same time preserving the relevant polarimetric information. The classification presented in this work involves two steps: first determine the reduced set of variables, from the data, that "optimally" represents the polarimetric information. Second, employ these newfound variables to segment (and classify) the polarimetric SAR image. Finally, the classification results are compared with standard statistical maximum likelihood techniques.