[Back]

M. Recheis:
"Automatic Recognition of Repeating Patterns in Rectified Facade Images";
Supervisor: W. Purgathofer, P. Musialski; Institut für Computergraphik und Algorithmen, 2009; final examination: 2009-12-16.

Building facades typically consist of multiple similar tiles which are arranged quite strictly in grid-like structures. The proposed method takes advantage of translational symmetries and is able to analyze and segment facades into tiles assuming that there are horizontal and vertical repetitions of similar tiles. In order to solve this quite complex computer vision task efficiently a Monte Carlo approach is presented which samples only selected image features. This method, which is meant to be a preprocessing step for more sophisticated tile segmentation and window identification in urban reconstruction tasks, is able to robustly identify orthogonal repetitive patterns on rectified facade images even if they are partially occluded, shadowed, blurry or otherwise damaged. Additionally, the algorithm is very running time efficient because neither quality of results nor the computational complexity are significantly depending on the image size.

http://www.cg.tuwien.ac.at/research/publications/2009/recheis-2009-arr/