期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2005
卷号：XXXVI-3/W24
出版社：Copernicus Publications
摘要：The spatial resolution of available image data plays an important role at the creation of object models for road extraction. The type and perceptibility of roads changes with increasing ground pixel size. The design of the model for the extraction of roads therefore has to be influenced by the resolution of the available imagery. In this paper we present a concept to automatically adapt road models for high resolution images to models suitable for images of lower resolution with similar spectral characteristics. The road model is formulated as a semantic net. Starting from the manually created semantic net for high resolution images and the given target scale, the road model is first automatically decomposed into groups of object parts. The representation of the object part groups in the coarser scale is then automatically predicted by scale change models, which are generated by deploying analytical as well as simulation procedures. The adapted object parts are at last fused back to a complete road model, which is suitable for road extraction in images of the lower target resolution. The automatic adaptation of a semantic net to a coarser scale is demonstrated for a given model for road extraction. The presented adaptation methodology facilitates the creation of new models for automatic object extraction in lower resolution images
