A large-scale evaluation of features for automatic detection of oil spills in ERS SAR images

The authors study the performance of automatic methods for oil spill detection in ERS SAR images. The presented algorithm has three main parts: (i) detection of dark spots; (ii) feature extraction; and (iii) dark spot classification. The dark spot detection locates all spots which can possibly be oil slicks in the image. For each slick, a set of backscatter, textural, and geometrical features are extracted. The dark spots are then classified into possible oil slicks and look-alikes based on the extracted features. Based on the current study, the authors believe that a semi-automatic oil slick identification system which can discriminate between oil slicks and look-alikes can be developed. To achieve this, some new features describing the surroundings of a slick and the slick itself must be defined and tested.