Facade repetition detection in a fronto-parallel view with fiducial lines extraction

Detecting repetitive structures on building facades plays an important role in facade image analysis. Observing that repetitions are usually horizontally and vertically aligned, and thereby can be localized by the horizontal and vertical lines passing along the repetition boundaries, we propose to detect repetitions by extracting these fiducial lines. Firstly, candidate lines are detected, containing both the fiducial lines and some mistaken lines passing across facade wall or repetitive structures. Secondly, to pick out the fiducial lines, we formulate a maximum a posterior problem to measure the probabilities that the lines can localize the repetitions. Finally, a dynamic programming based algorithm is developed to solve the problem efficiently. To evaluate the proposed approach, we implement a series of experiments on a dataset containing 60 facade images as well as the public Ecole Central Paris facade dataset. Both qualitative and quantitative results demonstrate the effectiveness of our approach.