A Robust Nonlocal Fuzzy Clustering Algorithm With Between-Cluster Separation Measure for SAR Image Segmentation

Fuzzy c-means (FCM) algorithm has been widely used in image segmentation, and there have been many improved algorithms proposed. But when dealing with synthetic aperture radar (SAR) images, they may not give satisfactory segmentation results because of speckle noise. In order to segment SAR image effectively, a robust Fuzzy clustering algorithm is proposed, called nonlocal fuzzy clustering algorithm with between-cluster separation measure (NS_FCM). In NS_FCM, to reduce the effects of the noise, we incorporate the nonlocal spatial information obtained using an improved nonlocal mean method, which adopts adaptive binary weighted distance measure and adaptive filtering degree parameter. In addition, we introduce a fuzzy between-cluster variation term into the objective function. Based on this, while minimizing the objective function, we can maximize the within-cluster compactness measure and the between-cluster separation measure of the partition simultaneously. Besides, by regulating the parameter of the fuzzy between-cluster variation term, we can adjust the distance between the clustering centers flexibly. This makes NS_FCM more effective to the images, which have some close classes in feature space. Experiments on synthetic and real SAR images show that the proposed method behaves well in SAR image segmentation performance.