Extracting Roads from Ikonos Images

Remote sensing has been used mostly to produce analyses on different aspects and phenomena concerning territory. The approach is not different from the one employed when analysing maps in the sense that analysis depends on the nature of data, maps and satellite images being two different sources of spatial data. Obviously, analysis based on cartographic data is different from the one based on satellite data, because of the different nature of data and the consequently different tools of analysis. Until now remote sensing techniques are not used in the mapping production, which rather makes use of photogrammetry in order to obtain all the cartographic objects defined by the cartographic legend. Research in applying image processing techniques to extract automatically cartographic features by using digital aerial photographs, as well as studies in integrating results from SPOT images with cartographic layers has been made in different case studies. The interest of this kind of research lies in the possibility of updating by automatic procedures the cartographic Gis. The themes mostly investigated are roads and urban areas. Following this line of research, the recent availability of Ikonos images at 1 meter's resolution leads researchers to put forward and experiment procedures to extract cartographic features in order to update Gis. It is worth noting that working in the cartographic context is quite different from working with analyses as in this case a specific cartographic feature must be distinguish and extract. The paper describes a procedure to extract main roads. The procedure has been designed for Ikonos multispectral images at 1 meter of resolution and applied in an area of the Venetian region bordering the Lagoon of Venice. In this territory we have to distinguish main roads from canals and country roads. The procedure can be divided into five phases: ! Extraction of linear elements ! Classification in order to extract pixels belonging to roads. ! Identifying elements common to the two previous results. ! Reconstruction of all roads and identifying/eliminating noises. ! Skeletoning, vectorialization, selection and reconstruction of roads. This work is part of a broader research line (MIUR-Cofin 2001) whose objective is to explore the possibility to extract road features from high-resolution images.