Abstract: Because of the influence of the speckle in the synthetic aperture radar (SAR) image, both the statistical dependencies among neighboring pixels and spatial adaptation of the clustering prototype should be considered during the process of SAR image segmentation. So it is required that the membership and spatial information should be combined. The adaptive spatial neighbor weighted possibilistic c-means (PCM) clustering based on the direction flow field is proposed in the paper. The direction flow field is constructed by combining the predictive coding model with the visual character and steerable wavelet transform. The relationship between the pixel under test and the pixels of neighborhood is described through Markov random fields (MRF) based on the direction flow field. Because the context information is considered, membership is adjusted efficiently. The experimental results on real SAR images demonstrate the merit of the proposed method, especially in despeckling and the preservation of details within a SAR image.

Key words: synthetic aperture radar (SAR) image segmentation, possibilistic c-means clustering, steerable wavelet transform, direction flow field