Contextual approach for oil spill detection in SAR images using image fusion and markov random fields

This paper presents a study for oil spill detection. The scheme incorporates contextual information using multi-conexity analysis. The image is modeled as a discrete Markov random field (MRF). Each pixel can be classified in two classes: {oil, not-oil}. To determine the class we optimized the a posteriori energy function by means of simulated annealing. The segmentation result contains different levels of information. In order to improve the detection, we propose a data fusion stage. To realize the data fusion we use a contextual algorithm. The result obtained is binary and shows in detail the oil spill in the analysis zone.