Model-based Segmentation Techniques for Multi-frequency Polarimetric Sar

A new technique, named DPOL MUM, is proposed for the segmentation of multi-frequency polarimetric SAR images. By following a Maximum Likelihood approach, an optimal split-merge test is derived, which fully exploits the characteristic block-diagonal structure of the data covariance matrix. This test is demonstrated to yield a reduced fluctuation error than the extension of the polarimetric test to the multi-frequency case (POL MUM). Moreover, such test represents the basic step of a presently implemented region-growing segmentation technique. The performance comparison shows that the newly introduced technique outperforms the previous split-merge techniques in the segmentation of synthetic and real multi-frequency polarimetric SAR images of land areas.