A hierarchical method for accurate water region detection in SAR images

Water region detection in SAR image is one of the significant applications for early warning of flood disaster. In this paper, a hierarchical method is proposed to detect water region in SAR images. The detection is a hierarchical process, and can be divided into two main stages: Bag of Visual words (BOV) representation based candidate region extracting, and false-alarm elimination based on topological characteristic description in gray level. The experiments based on SAR images of real complex scenery show that the proposed method achieves higher accuracy and robustness. Particularly, the topological features can eliminate the false-alarms caused by mountains, buildings, vegetation and other shadows.