A Non-Local Fuzzy C-Means Clustering Segmentation Algorithm Based on Comentropy and Between-Cluster Scatter Matrix to Overcome the Inherent Coherence Speckles of SAR Images

Abstract

The fuzzy c-means (FCM) algorithm and many of its variations have been widely adopted for image segmentation tasks. However, these methods are unable to present satisfactory segmentation results when dealing with synthetic aperture radar (SAR) images owing to the intrinsic speckle noise. In order to achieve the effective segmentation of SAR images, a robust FCM algorithm, namely NCBS_FCM, is proposed. The nonlocal spatial information is utilized to reduce the effect of speckle noise. Furthermore, NCBS_FCM takes advantage of the comentropy based on local gray histogram to acquire the adaptive weighting parameter for nonlocal spatial information term, which can achieve a better balance between speckle suppression and edge detail preservation. In addition, this paper incorporates the between-cluster scatter term into the objective function to adjust the distance between the cluster centers accordingly. Therefore, NCBS_FCM is more robust to various images and achieves satisfactory segmentation accuracy. Experiments on simulated and real SAR images show that NCBS_FCM outperforms other proposed variations of FCM by a significant margin.