A Hierarchical Segmentation Method for Multispectral Imagery

GUO Jiancong,LI Peijun,XIAO Xiaobai
DOI: https://doi.org/10.3321/j.issn:0479-8023.2009.02.019
2008-01-01
Abstract:A hierarchical segmentation method is proposed for multispectral imagery by combining extended watershed transform and dynamics of contours. First, the gradient magnitude of an image is computed, and then local irrelevant minima are eliminated. After computing dynamics of contours, hierarchical (multi-level) segmentation results are produced by thresholding. The proposed method is evaluated with a high spatial resolution image of the urban area, and compared with the existing image segmentation method. The results show that the method can get hierarchical segmentation results, and segmente most of image objects. The segmentation results give the hierarchical representation of land cover types, which is helpful for further object based analysis.
What problem does this paper attempt to address?