A Topology-Based Architectural Search Engine

Architectural design involves the study and manipulation of complex spatial hierarchies and relations. Current search tools such as Google Image and websites like Pinterest are based on text-input from users, or use algorithms to determine visual similarity, and thus their results are not useful to explore spatial relations. There is thus a gap between the logic of raster or tag based search engines and the demands of architectural users. My thesis helps narrow down this gap by introducing a novel architectural search engine based on spatial relations. The system offers users a visual tool to depict the spatial structure of an architectural plan, which it uses to generate a graph of the building. The system then uses this information to quantify the degree of spatial similarity of the building in comparison with examples in a database, using a belief propagation algorithm. The thesis discusses the potential of this system to elicit new ways of searching architectural information and to scale into an open platform where users add new diagrams of existing and new designs.