Outlier Rejection by Means of Median Matrices for Polarimetric SAR Covariance Symmetry Classification

This letter exploits the intrinsic selectivity properties of the median to enhance the covariance symmetry classification in polarimetric synthetic aperture radar (PolSAR) images. More in detail, the median matrices are utilized to properly detect and remove outliers in the data belonging to a reference window, in turn used to estimate the covariance structure of the pixel under test. Hence, the scene is classified in terms of the structures assumed by the covariance under specific symmetric scattering mechanisms. To do this, for each pixel under test, the data in a reference window are filtered through the application of a generalized inner product (GIP)-based procedure involving the median matrix in its computation. The filtered data are then used as input to a model order selection (MOS)-based procedure for the final scene classification. Tests conducted on L-band real-recorded SAR data show the effectiveness of the devised framework.