Abstract: In the face of massive data of high-resolution SAR images, the academic community widely simplifies image analysis processing through the superpixel-based approach. The superpixel segmentation algorithm which is generally suitable for optical images is not ideal for SAR images with speckle noise. Improving the existing superpixel generation algorithms for SAR image has been a hot topic among the scholars. In this paper, we discussed the feasibility of introducing edge strength feature into the superpixel segmentation algorithm. By combining the edge strength feature with the linear spectral clustering method, a novel superpixel generation algorithm (e-LSC) for SAR image was proposed. Compared with several typical superpixel generation algorithms on the simulated SAR image and the real SAR image, it is verified that the segmentation performance of e-LSC algorithm on the boundary adherence and the regularization of the homogeneous area is improved.

Key words:  synthetic aperture radar, superpixel, linear spectral cluster, edge strength feature