SAR image segmentation using B-Spline deformable contours

Synthetic Aperture Radar (SAR) images are corrupted by a signal-dependent non-additive noise called speckle. Many statistical models have been proposed to describe this noise, aiming at the development of specialized techniques for image improvement and analysis. One of the most important parameters in SAR imagery is texture or roughness that, within some statistical models, can be characterized by a scalar. This quantity is obscured by speckle noise. The G distribution is a quite exible model that succeeds in describing areas with a wide range of roughness, from pastures (homogeneous) to urban areas (extremely heterogeneous). This distribution exhibits a remarkably good performance within urban areas, while other distributions considered in the literature for SAR data, namely Gamma and K, fail to t that type of data. In addition to its expressiveness, a sub-case of the G distribution, the G0 distribution is mathematically more tractable than the classical K law. These parameters will be estimated in order nd the transition points between regions with di erent degrees of homogeneity. In order to determine the boundaries of urban areas in SAR imagery B-Splines is here proposed. After the speci cation of an initial region within the city to be segmented, the algorithm determines the positions of the B-Spline control points maximizing an objective function. The proposed algorithm is tested on synthetic SAR images in order to measure its performance. Eje: Imágenes Red de Universidades con Carreras en Informática (RedUNCI)