A Study Of SAR Image Segmentation Based On Fuzzy C-Means Clustering Algorithm With Neighborhood Information

High-resolution Synthetic Aperture Radar(SAR) has been attracted a wide spread attention over the years, the range resolution and azimuth resolution of SAR has been improved by the techniques called pulse compression technology and synthetic aperture principle, which makes it possible that we can get large areas of high resolution radar images. What's more, SAR has the ability to detect objects and reconnaissance without interruption and to detect objects from stand ranges in bad weather. SAR images include plenty of amplitude, phase and polarization information of targets, which offsets the deficiency of optical images.SAR image segmentation is a key part of SAR image interpretation. For the lack of valuable prior knowledge of speckle noise, SAR image segmentation tends to use unsupervised segmentation methods. Fuzzy C-Means clustering algorithm(FCM) is a kind of unsupervised segmentation methods and it receives widespread attention. The fuzzy set theory used in FCM can settle the problem in SAR image segmentation caused by partial information or contradiction. FCM has obtained the predictive effect when applied it on optical image, but not on SAR image. Speckle noise in SAR image increase the difficulty of SAR image segmentation. In this paper, we propose two variants of FCM by introducing new constraint items, which is built for speckle noise. The algorithms are presented below.(1) Proposing a robust self-adaption fuzzy c-means algorithm for SAR image segmentation(RSFCM), firstly, we make an analysis on some algorithms based on FCM and found that though it has some effect on cutting down the effect of speckle noise by taking advantage of neighborhood information in the process of clustering, the misclassification phenomenon appears easily for the noise interference. To reduce the rate of misclassification, a tradeoff weighted factor is introduced into RSFCM. The tradeoff weighted factor can adaptively change with the degree of pixel corrupted by speckle noise to cut down noise interference. Secondly, we can get the objective function of RSFCM by introducing the tradeoff weighted factor. Thirdly, experiments on synthetic SAR image and real SAR image shows that RSFCM can reduce the rate of misclassification and can raise the accuracy of segmentation result.(2) Proposing a fuzzy c-means clustering algorithm incorporating neighborhood relations for synthetic aperture radar image segmentation. At first, to estimate the degree of noise influence on image pixels, we extract the target information based on the principle of probability maximization and the spatial correlation between neighboring pixels. The extraction of target information considers the features of speckle noise, which increase the reliability of target information. And then, we build an adaptive parameter incorporating neighborhood relations to balance the relations between noise and image details. At last, we get the objective function of FCMNR by introducing the adaptively parameter. Experiments performed on SAR images illustrate the excellent performance of the new method.