Bayesian Segmentation of Oceanic SAR Images: Application to Oil Spill Detection

摘要

This paper introduces Bayesian supervised and unsupervised segmentationalgorithms aimed at oceanic segmentation of SAR images. The data term,\emph{i.e}., the density of the observed backscattered signal given the region,is modeled by a finite mixture of Gamma densities with a given predefinednumber of components. To estimate the parameters of the class conditionaldensities, a new expectation maximization algorithm was developed. The prior isa multi-level logistic Markov random field enforcing local continuity in astatistical sense. The smoothness parameter controlling the degree ofhomogeneity imposed on the scene is automatically estimated, by computing theevidence with loopy belief propagation; the classical coding and least squaresfit methods are also considered. The maximum a posteriori segmentation iscomputed efficiently by means of recent graph-cut techniques, namely the$\alpha$-Expansion algorithm that extends the methodology to an optional numberof classes. The effectiveness of the proposed approaches is illustrated withsimulated images and real ERS and Envisat scenes containing oil spills.