SAR image segmentation with active contours and level sets

Automatic interpretation of synthetic aperture radar (SAR) images requires automatic segmentation of these images. Image segmentation is a fundamental problem in computer vision, particularly difficult with SAR images because of the presence of strong, multiplicative speckle noise. The purpose of this study is to investigate a novel algorithm for segmenting a synthetic aperture radar (SAR) image into a fixed but arbitrary number of Gamma-homogeneous regions. This unsupervised algorithm is based on active contours and consists in evolving closed simple planar curves to minimize a criterion containing a term of conformity of data to a model of SAR image intensity and a term of regularization. The curve evolution equations are implemented via level sets for numerical stability and to allow variations in the topology of the curves during their evolution. Examples are given using real SAR images.