Abstract

We present an approach for detecting and matchingbuilding facades between aerial view and street-view im-ages. We exploit the regularity of urban scene facadesas captured by their lattice structures and deduced frommedian-tiles’ shape context, color, texture and spatial sim-ilarities. Our experimental results demonstrate effectivematching of oblique and partially-occluded facades be-tween aerial and ground views. Quantitative comparisonsfor automated urban scene facade matching from threecities show superior performance of our method over base-line SIFT, Root-SIFT and the more sophisticated Scale-Selective Self-Similarity and Binary Coherent Edge descrip-tors. We also illustrate regularity-based applications ofocclusion removal from street views and higher-resolutiontexture-replacement in aerial views.