Feature shape and elevation based road classification and extraction on high spatial resolution remote sensing imageries

Classification and extraction of geospatial features from high spatial resolution imageries approved is one of the most significant steps for spatial database acquisition and updating in GIS. This research is to explore the methodologies of recognizing shape and elevation characteristics of spatial features on the remote sensed images. We focus on the road network classification and extraction among various features on the ground because it carries unique characteristic of elongation. We combined both a pattern recognition model of connected component labeling (CCL) in two dimensional image processing, and a three dimensional DEM elevation filtering model to extract the road and street features from high resolution imageries. Samples of digital aerial photographs in Erie County, New York were used to test the methodology. The results indicated that the correctness of road extraction in rural areas can reach 69.1%; that of completeness is 74.9%; and the overall quality is 73.1%. By contrast, the correctness in urban high-rise region is only 39.5%; that of completeness is 42.8%; and the overall quality is 32.8%.