Segmentation Based on Normalized Cuts for the Detection of Suburban Roads in Aerial Imagery

This paper deals with the segmentation of images of suburban scenes with the normalized cut algorithm. The segmentation results are intended to be used for the extraction of roads in order to assess existing road data. The similarity matrix necessary for the normalized cuts algorithm is built up using similarity criteria that are suitable for the separation of road segments and non-road segments. These criteria are edges, colour, hue and road surface colour derived with the help of the database information which is thus used as prior information to facilitate the segmentation and extraction. Segmentation is the main topic of this paper, but some hints on future work regarding the selection of road segments based on road colour are given. The results show that the approach is suitable for the segmentation in order to extract roads in suburban scenes.