Mixture-Based Superpixel Segmentation and Classification of SAR Images

We propose a mixture-based superpixel segmentation method for synthetic aperture radar (SAR) images. The method uses SAR image amplitudes and pixel coordinates as features. The feature vectors are modeled statistically by taking into account the SAR image statistics. We resort to finite mixture models to cluster the pixels into superpixels. After superpixel segmentation, we classify different land covers such as urban, land, and lake using the features extracted from each superpixel. Based on the classification results obtained on real TerraSAR-X images, it is shown that the results obtained by the proposed superpixel method are capable of achieving a more accurate classification compared with those obtained by state-of-the-art superpixel segmentation methods such as quick-shift, turbo pixels, simple linear iterative clustering, and pixel intensity and location similarity.