As the reflectivity is very low in near-infrared spectral band for water, it is easy to extract water bodies in remote sensing images with the use of NDVI (Normalized Difference Vegetation Index).The problem is that we always have to try several times to find the appropriate threshold to separate water bodies from land. In this paper, a particular method was developed to solve this problem by automatically determining the threshold that was used to extract the specific water body. We select both Chaohu Lake and Taihu Lake as the study regions. First, we generate NDVI image of the study region from GF-1 data after several pre-processing procedures. Then we resize the NDVI image to make it contain approximately the same number of water and land pixels. Because the NDVI value is lower for water than that for land, there will appear two peaks in the histogram which we derived from the resized NDVI image. The threshold locates at the lowest place between the two peaks can be chosen as the proper threshold used for land/water delineation. With the use of a reasonable threshold range we can finally get the threshold by calculating the minimum value in it, and extract the water body successfully.

Date Published: 10 December 2014
PDF: 7 pages
Proc. SPIE 9261, Ocean Remote Sensing and Monitoring from Space, 926103 (10 December 2014); doi: 10.1117/12.2068815

Published in SPIE Proceedings Vol. 9261:
Ocean Remote Sensing and Monitoring from Space
Robert J. Frouin; Delu Pan; Hiroshi Murakami, Editor(s)