Comparison of segmentation methodologies applied to ERS-1 SAR images

The performance of three segmentation methods, developed for high resolution airborne synthetic aperture radar (SAR) data, on lower resolution spaceborne data is assessed using two ERS-1 images. Methods based on a cartoon model for the underlying image are shown to be able to successfully segment regions bounded by abrupt edges, but to be unable to represent gradual changes in intensity and linear features. These problem are overcome at the expense of greatly increased cpu time using MAP reconstruction which models the underlying image as a semi-continuous surface