Detecting Large Repetitive Structures with Salient Boundaries Changchang Wu 1 , Jan-Michael Frahm 1 , and Marc Pollefeys 2 1 Department of Computer Science UNC Chapel Hill, NC, USA {ccwu,jmf}@cs.unc.edu 2 Department of Computer Science ETH Z¨ urich, Switzerland marc.pollefeys@inf.ethz.ch Abstract. This paper presents a novel robust and efficient framework to analyze large repetitive structures in urban scenes. A particular con- tribution of the proposed approach is that it finds the salient boundaries of the repeating elements even when the repetition exists along only one direction. A perspective image is rectified based on vanishing points computed jointly from edges and repeated features detected in the orig- inal image by maximizing its overall symmetry. Then a feature-based method is used to extract hypotheses of repetition and symmetry from the rectified image, and initial repetition regions are obtained from the supporting features of each repetition interval. To maximize the local symmetry of each element, their boundaries along the repetition direc- tion are determined from the repetition of local symmetry axes. For any image patch, we define its repetition quality for each repetition interval conditionally with a suppression of integer multiples of repetition inter- vals. We determine the boundary along the non-repeating direction by finding strong decreases of the repetition quality. Experiments demon- strate the robustness and repeatability of our repetition detection. 1 Introduction Repetition and symmetry are frequently used in the design of urban architecture. In fact, buildings often consist of a hierarchy of repetitions and symmetries (e.g. Fig. 1). Particularly, most of the basic repeating elements on facades (such as doors and windows) are symmetric by themselves, repetition and symmetry coexist and interplay at different scales. This paper introduces a new method to detect repeating elements with salient boundaries in facade images. The symmetry and repetition patterns together with the appearance of the repeating/symmetric elements provide a strong characterization of the scene. Given that, particularly for urban scenes, the symmetries and repetitions of a scene describe its high-level structure, they can be used for wide baseline match- ing. One area where this representation would be useful is in the reconstruction from urban photo collections as in [1]. The reliable boundaries of the detected repeating elements and the symmetric structure can be used as compact image K. Daniilidis, P. Maragos, N. Paragios (Eds.): ECCV 2010, Part II, LNCS 6312, pp. 142–155, 2010. c  Springer-Verlag Berlin Heidelberg 2010