SAR image segmentation based on spatially adaptive weighted possibilistic c-means clustering

Abstract

Due to the influence of speckle in synthetic aperture radar (SAR) image, statistical dependencies among neighboring pixels should be considered in SAR image segmentation. The spatially adaptive weighted possibilistic c-means (SAW-PCM) clustering algorithm is proposed in which spatial information is introduced into PCM approach to directly adjust the membership. The relationship between the neighboring pixels is described through Markov random fields (MRF). To preserve detail information in SAR images, the directional neighborhood system set is established. The selection of neighborhood systems is based on similarity measurement (SM) between wavelet energies of comprehensive result of steerable wavelet transform. Among the different neighborhood alternatives, the one with the highest SM value is chosen to compute the weight value. The experimental results on real SAR images demonstrate the merit of the proposed method, especially in the preservation of details within a SAR image.