Superpixel Segmentation for PolSAR Images Based on Geodesic Distance

Abstract

Considering the lack of similarity capabilities of the distance metric used in the traditional Polarimetric Synthetic Aperture Radar (PolSAR) image superpixel segmentation algorithm, a novel PolSAR image superpixel segmentation algorithm based on geodesic distance is proposed in this paper. First, the PolSAR image is initialized as a hexagonal distribution, and all pixels are initialized as unstable pixels. Thereafter, the geodesic distance between two real symmetric Kennaugh matrices is used to measure the similarity between the current unstable point and another cluster point in the search region to more accurately assign labels to unstable points, thereby effectively reducing the number of unstable points. Finally, the postprocessing procedure is used to remove small, isolated regions and generate the final superpixels. To verify the effectiveness of the initialization method and the high efficiency of the geodesic distance, extensive experiments are conducted using simulated PolSAR images. Moreover, the proposed algorithm is analyzed and compared with four other algorithms using simulated and real-world images. Experimental results show that the superpixels generated using the proposed method exhibit higher computational efficiency and a more regular shape that can more accurately fit the edges of real objects compared with those using the four other algorithms.