Variation-Factored Encoding of Facade Images

Abstract

Urban facades contain large-scale repetitions in the form of windows, doors, etc. Such elements often are in different configurations (e.g., open or closed) obscuring their regular arrangements to any direct low-level pixel matching based repetition detection.We propose a variation-factored representation for facade images by progressively favoring larger repeated structures while allowing relabeling using candidate element types. We formulate the problem as a Markov Random Field (MRF) based optimization, and evaluate the algorithm on a large number of benchmark facade images. Such a facade encoding is very compact and can be used for rapid generation of realistic 3D models with variations suitable for online map viewers or mobile navigation aids.