Data caption and conversion are two of the most costly operations of any GIS, in terms of computer time and manual work needed for spatial data acquisition. They can represent up to 80 percent of the total implementation costs. Manual digitising is a very error prone and costly operation, especially due to the lack of explicit topology in commercial GIS systems. Indeed, each map update might require the batch processing of the whole map. Currently, commercial GIS do not offer completely automatic raster/vector conversion even for simple scanned black and white maps. Various commercial raster/vector conversion products exist for the skeletonisation or thinning of the pixels forming the line, but these approaches have shown difficulties with the extraction of good topology. The spatial feature extraction in raster/vector conversion systems is based on line tracing algorithms. In order to operate they need user defined tolerances settings, what causes difficulties in the extraction of complex spatial features, for example: road junctions, curved or irregular lines and complex intersections of linear features. The approach we use here is based on image processing filtering techniques to extract the basic spatial features from raster data. These spatial features can be used for the reconstruction of the image within the topological data structure - the Voronoi diagram. The novel part of this research is the definition of deterministic topological rules and algorithms for extracting the spatial features from the Voronoi data structure. These spatial features can then be represented in different spatial data structures that can be implemented in a GIS. In this research we use the topological approach to develop new algorithms and data structures for integrated raster/vector models leading to the improvement of data caption and conversion in GIS and to develop a software toolkit for automated raster/vector conversion. The approach is based on computing the skeleton from Voronoi diagrams using natural neighbour interpolation. In this paper we present the algorithm for skeleton extraction from scanned maps. We show that the skeleton extracted from the map features can approximate the centreline of the map object. We apply this algorithm directly on the Voronoi cells, for the extraction of complex spatial features. This research can lead to the improvement of current practices in spatial data acquisition reducing significantly the cost and amount of work needed.
Proceedings of the 20th Isprs Congress: Geo-imagery Bridging Continents, 2004