Synthetic aperture radar image segmentation based on well-initialized active contours

For the presence of speckles, synthetic aperture radar (SAR) image segmentation is often acknowledged as a difficult problem. Large efforts have been done to cope with the influence of speckles on image segmentation. However, it is still an open problem. This paper proposes a method based on well-initialized active contours of the Chan–Vese model. First, Gabor filter banks are used to efficiently suppress the speckles, which simultaneously modify distributions into the ones with several local symmetrical peaks. Then, such modified distributions are described with Gaussian mixture models, the parameters of which are estimated using the expectation maximization algorithm. Third, pixels in filtered images are classified according to their probability of belonging to each Gaussian distribution, and the results are used to initialize the active contours of the Chan–Vese model, which are conducted on the unfiltered SAR images. Experimental results with simulated images and real SAR images validated the feasibility and effectiveness of the proposed segmentation method.