A Semantic Region Growing Algorithm: Extraction of Urban Settings

Recent years have witnessed a growing production of Volunteer Geographic Information (VGI). This led to the general availability of semantically rich datasets, allowing for novel ways to understand, analyze or generalize urban areas. This paper presents an approach that exploits this semantic richness to extract urban settings, i.e., conceptually-uniform geographic areas with respect to certain activities. We argue that urban settings are a more accurate way of generalizing cities, since it more closely models human sense-making of urban spaces. To this end, we formalized and implemented a semantic region growing algorithm—a modification of a standard image segmentation procedure. To evaluate our approach, shopping areas of two European capital cities (Vienna and London) were extracted from an OpenStreetMap dataset. Finally, we explored the use of our approach to search for urban settings (e.g., shopping areas) in one city, that are similar to a setting in another.