Superpixel Generation for PoISAR Images with Global Weighted Least-Squares Filtering and Linear Spectral Clustering

Most traditional superpixel generation methods for polarimetric synthetic aperture radar (PoISAR) images focus on the accurate representation of data similarity between pixels, which, however, is often offset by the spatial distance and also suffers from the speckle noise that is less considered. To release this problem, this paper proposes a novel superpixel generation approach for PoISAR images based on global weighted least squares (GWLS) filtering and linear spectral clustering (LSC). 'the method consists of three modular steps. Firstly, a GWLS filter is employed to suppress the speckle in the PoISAR image Next, a pseudo-color image is constructed with the three components of the filtered Pauli scattering vector. Finally, the LSC method is selected to process the pseudo-color image and then yield the final superpixels of the PoISAR image. The proposed algorithm is clear in structure and simple in implementation. The comparison experiment results performed on actual PoISAR images have validated the effectiveness of the proposed method.