Extracting of urban features from high resolution remote sensing data based on multiscale segmentation

A multiscale segmentation method is proposed for multispectral imagery of high resolution by combining an adapted watershed algorithm and a region merging algorithm. Before the preliminary segmentation by the adapted watershed algorithm, a filtering method and a method for getting rid of local minimum areas are imposed to avoid over-segmentation. The whole process can be divided into five steps as follows. A case study is conducted with a high resolution image, QuikBird, of Beijing city acquired in 2007. From the segmentation results it can be found most of urban features could be extracted correctly and the segmentation edge is accurate and smooth. And it can be concluded that the method can have more semantic information, reduce the dasiaPepper and Salt Phenomenonpsila effectively, and improve the overall classification accuracy of QuikBird image with improved computing efficiency.