In this thesis, we present a two-stage framework for extracting what we define as a quasi-regular structure in facade images. A quasi-regular structure is a rectangular grid representing the placements of repetitive architectural objects, for example, windows in a facade. Such a structure generalizes a perfect single lattice structure with manageable complexity. In our two-stage framework we first propose to formulate window detection using an object-oriented Marked Point Process model with an efficient MCMC sampler. This leads to an initial structure map which indicates potential window locations. Then in the second stage, we propose a regularization scheme based on the initial structure map to recover the complete structure. This stage takes advantage of previously obtained...[ Read more ]

In this thesis, we present a two-stage framework for extracting what we define as a quasi-regular structure in facade images. A quasi-regular structure is a rectangular grid representing the placements of repetitive architectural objects, for example, windows in a facade. Such a structure generalizes a perfect single lattice structure with manageable complexity. In our two-stage framework we first propose to formulate window detection using an object-oriented Marked Point Process model with an efficient MCMC sampler. This leads to an initial structure map which indicates potential window locations. Then in the second stage, we propose a regularization scheme based on the initial structure map to recover the complete structure. This stage takes advantage of previously obtained object information and the intrinsic low rank constraint of the quasi-regular structure and can thus reliably extract underlying structures from arbitrary shaped windows. We have extensively evaluated our method over a large variety of facade images, and demonstrated both the efficiency and robustness of our two-stage framework. In addition, both qualitative and quantitative comparisons with other state-of-the-art methods are presented to show the superiority of our framework.