Water Body Automatic Extraction from Remotedly Sensed Images

摘要

In this paper, we put forward an algorithm to automatically extract the water body information from satellite and airborne images. Water is often one of the dominating ground targets in remote sensing images. And it is important that we can extract the water-land boundary quickly and with high resolution.Some research efforts have been made on this issue in the last decade. Du (Du et al, 1998) developed a knowledge-based model to extract water information from NOAA/AVHRR. Wan (Wan et al, 2000) presented a method based on seed points and connectivity analysis. Most recently, Qin (Qin et al, 2001) put forward a multi-spectral classification-based algorithm for water extraction. However, the image processing method involves some artificial participance, while the feature classification method is somewhat time-consuming, because we need the water information only. In this paper, we use image segmentation and multi-spectral classification for automatic water body extraction. Our algorithm based on the prerequisite that water has distinct spectral responding from all the other ground objects. First, we use image segmentation to separate the water body from the background, which involves the use of histogram analysis. We can obtain the pixel value information and the approximate locations of water body, such as lakes, rivers. The information is then used to search out the water regions by seed-point-filling and multi-spectral identification techniques. Extensive experiments were carried out on Landsat-7 ETM Plus, SPOT, IKONOS, and aero photographs, etc. We get encouraging results when apply the method in all kinds of regions, including ocean, coast, rivers, lakes.