Abstract
We propose a novel regularity-driven framework for fa-cade detection from aerial images of urban scenes. Gini-index is used in our work to form an edge-based regularitymetric relating regularity and distribution sparsity. Facaderegions are chosen so that these local regularities are max-imized. We apply a greedy adaptive region expansion pro-cedure for facade region detection and growing, followedby integer quadratic programming for removing overlap-ping facades to optimize facade coverage. Our algorithmcan handle images that have wide viewing angles and contain more than 200 facades per image. The experimentalresults on images from three different cities (NYC, Rome,San-Francisco) demonstrate superior performance on fa-cade detection in both accuracy and speed over state of theart methods. We also show an application of our facadedetection for effective cross-view facade matching.