EIA 2006 |
Society for Environmental Information Sciences
Environmental Informatics Archives
ISSN 1811-0231 /
ISEIS Publication Series Number P002
2006 ISEIS. All
Paper EIA06-034, Volume 4
(2006), Pages 370-383
Image Segmentation for classification of land covers based on high spatial resolution sensor imagery
H. Lin1,2, D. K. Mo1, H. Sun1 and Y. J. Xiong1
1. The University of Central South Forestry & Technology, Changsha, 410004 Hunan, P. R. China.
2. The University of Zhejiang, Hangzhou, 310000 Zhejiang, P. R. China.
Image segmentation is a process of delineating an image into homogeneous polygons related to objects on the ground, and it is the foundation for further image analysis and interpretation. Low- or medium-resolution remotely sensed imagery usually leads to low accuracy of image segmentation because of large pixel sizes and a lot of mixed pixels. Thus, use of high-resolution imagery will probably result in increase of image segmentation accuracy because of smaller area covered by each pixel and reduced mixed pixels. This paper presents a study of QuickBird image segmentation for classification of land covers using mean-shift algorithm. The study area includes 1024 * 1024 pixels and covers the campus area of Central South Forestry University, Zhuzhou, Hunan, China. Six types of land cover were classified and the accuracy of classification was examined. The result showed that: the mean-shift algorithm led to a high accuracy of classification. Computing time for segmentation at different scales was also analyzed.
Keywords: High spatial resolution, Image Segmentation, Land cover, Mean-shift algorithm, QuickBird imagery
Full Text (PDF)