Context segmentation of oceanic SAR images: Application to oil spill detection

This paper introduces an algorithm based on the context of MRF (Markov random field) model, and this method achieved oil spilling detection and segmentation. In this paper, the two important elements are initial labelling field and potential parameter estimation. The algorithm model chooses optical pyramid of saliency map as initial label field and Ising model as segmentation function. Using the GMM (Gaussian Mixture Model) and MAP (Maximum a Posterior) get local optimal result by ICM (Iteration Condition Model) method. This paper is also deeply researching the potential parameter which is the impact factor in segmentation function. Through studying the relationship between potential function and every scale-levels of saliency pyramid, the paper gets the better result which is more accuracy segmentations and keeping more texture information. The series experiments prove this method having false alarming rejection and noise suppression function in Oceanic SAR images.