FDIE: a graph-based framework for extracting design information from annotated building facade images

Abstract

ABSTRACT Images of building facades, rich in design information, are important for architectural studies. In recent years, numerous studies have explored the automatic recognition and annotation of basic elements (e.g. windows) in these images. However, most application of the recognized elements remains at the pixel level, such as 3D reconstruction, and the design information hidden behind the elements cannot be processed using common data analysis methods because the annotations are still unstructured data. This prevents us from further discovering the composition pattern of building facades from the images. To enable the recognized elements to be analyzed, a framework for facade design information extraction (FDIE) from annotated building facade images based on graph theory is proposed in this paper. The design information is divided into three categories, i.e. graphic, relationship, and sum information, and extracted as structured data. An image restoration test demonstrates that more than 95% of the design information can be extracted using the FDIE. These structured data provide an important basis for intelligent analysis of composition pattern. Several examples and cases partly illustrate the application prospects of the extracted information. Other graph-based analysis methods are also expected to be applied to the extracted design information.