A Typology of Architectural and Graphic Primitives in Freehand Representations

9.1 The architect’s papers Despite being a mixed and transitional case (Goodman 1976), visual architectural representations have been extremely productive and expressive design tools. Drawings have replaced buildings as the primary product of the architect since the Renaissance, in what appears to be an attempt to elevate the status of the master builder to that of a scientific or artistic designer (Evans 1995; Porter 1997). This has resulted into not only many worthwhile buildings but also a huge number of images that document design processes and products from a variety of viewpoints. Our knowledge of world architecture relies more on published images than on personal experience (Evans 1989). In recent years our already highly visual culture has been further enhanced by social and technological developments, including the democratisation of the computer, the proliferation of the Internet and the popularisation related multimedia technologies. These have created an opportune environment for the digital dissemination of visual architectural information, uniquely suited as a repository of architectural imagery. In most current information systems images are treated as discrete, integral entities. They are indexed through textual annotations and associated with other information items manually. Analysis of images, e.g. in the framework of content-based retrieval, can recover a number of important features, such as the number and shape of spaces or the overall shape of a floor plan. It can also produce the general topological pattern of such features. For practical reasons, however, image analysis is generally restricted to a small number of salient features. These form the basic classification criteria for the image in the framework of specific activities. Therefore, image analysis becomes a deterministic search for the most economical collection of features (Gong 1998). The approach underlying our research is fundamentally opposed to both manual indexing and basic feature extraction, on the grounds that they do not return a formal representation, which could be considered equivalent to the image itself. Drawing from computer vision and image analysis we aim at a comprehensive recovery of primitive parts, identification of relationships between these parts and recognition of emerging structures.