Model-based Scattering Power Decomposition with Superpixels for the Coherency Matrix

Pixel-wised target decomposition methods ignore the neighborhood spatial information between pixels and do not use the consistency of local region. To overcome this problem, a model-based scattering powers decomposition with superpixels is proposed for POLSAR data in this letter. Simple linear iterative cluster (SLIC) is used to segment the polarimetric synthetic aperture radar (POLSAR) image into many small regions, i.e., superpixels, and Non-Local means filter for POLSAR image is used when the dominant scattering mechanism is preserved at the same time. The real POLSAR data on San Francisco are used to prove the efficacy of the proposed decomposition. Experiment results show that the scattering powers from the proposed decomposition are more effective to represent the fully POLSAR data.