Detecting dark spots from SAR intensity images by a point process with irregular geometry marks

Aiming at the difficulties to determine the geometric shape of dark spots, this paper proposes an Irregular Geometry Marked Point Process (IGMPP) to detect dark spots from Synthetic Aperture Radar (SAR) images. Firstly, a series of random points used to define the locations of dark spots in the SAR image domain, and an irregular polygon mark associated with each random point indicates the shape of each dark spot. Subsequently, the pixels intensities in and out of the polygons are characterized with independent and identical Gamma distributions, respectively. By Bayesian paradigm, the number, sites, and geometry parameters of polygons are modelled with a posterior distribution. In order to simulate the posterior, a Reversible Jump Markov Chain Monte Carlo (RJMCMC) strategy is developed. Then, the optimal parameters concerning the dark spots can be obtained by a Maximum A Posteriori (MAP) scheme. Experiments are performed by simulated SAR images. The experimental results show that the proposed algorithm is effectively and efficiently applied to detect the dark spots.