Coastline detection from SAR images

Coastline detection from synthetic aperture radar (SAR) images is in the category of edge or boundary detection. But the lack of contrast between the ocean and nearby land areas makes the ordinary edge detection methods inferior. Ocean areas in SAR images, however, are much more homogeneous in grey levels than land areas and features reflecting the “roughness” of an image can be very useful for sea-land separation. This paper tests several features. Experiment results show that the best result is given by the absolute deviation from the mean extracted using a 7×7 scanning window and that a reasonable sea-land separation can be obtained from it with the help of Gibbs distribution. But misclassification occurs if a land area is homogeneous enough or a sea area is also “rough”. This is manifested in the segmentation image as surplus small “lakes” and “islands” which are eliminated if their area is below the user-specified threshold. Another problem is the blurring effect of the window operation. The consequence is found to be an over-segmentation of the land areas and so deblurring is done through an even contraction of the land areas. The detected coastlines are found, when overlaid on the original image, to coincide quite well with the true situation