Multi-scale segmentation of the high resolution remote sensing image

The high spatial resolution remote sensing image provide more details such as color, size, shape, context and texture. The traditional pixel-based classifier cannot provide satisfying results and may reduce the classification accuracy. object-oriented processing for extraction of information from remote sensing data become of interest. The first step of the object-oriented processing is the segmentation. The object is derived by, means of multi-scale segmentation in this paper. The hierarchical image segmentation and region-merging are implemented. The procedure of the region-merging does not stop until the average size of all object regions exceeds the scale threshold. Lastly, in order to provide an appropriate link between remote sensing image and GIS data, the region boundary will he obtained from the segmented remote sensing image and initialize the regions' boundary to a series of polygon by Douglas-Peucker (DP) algorithm.