Abstract

We propose a sequential optimization technique for segmenting a rectifified image of a fac¸ade into semantic categories. Our method retrieves a parsing which respects common architectural constraints and also returns a certifificate for global optimality. Contrasting the suggested method, the considered fac¸ade labeling problem is typically tackled as a classifification task or as grammar parsing. Both approaches are not capable of fully exploiting the regularity of the problem. Therefore, our technique very signifificantly improves the accuracy compared to the state-of-the-art while being an order of magnitude faster. In addition, in 85% of the test images we obtain a certifificate for optimality