Segmenting Satellite Imagery: A Region Growing Scheme
Li Kai, Jan‐Peter Müller · 2005
A new approach, the LM method, to region growing image segmentation is described, in which regions grow one by one in all directions from seed points until the absolute difference of pixel values is above a local difference threshold and/or when there is an edge pixel. This approach has been implemented on the SUN workstation and a PARSYS SN1000 transputer array. A quantitative study to compare segmentation results with field information derived from 1:25,000 scale maps is described. The experiments reported here with LANDSAT TM and SPOT PA and XS data have shown that the new method is capable of extracting areal features from satellite imagery even with very low contrast and with multidimensional or single-band data sets.