In his book The Aesthetics of Townscape (1979), Yoshinobu Ashihara argues that one of the factors that determines the beauty of streetscapes is the form created by the individual building exteriors. He defines the contours created by the outlines of buildings along the street as primary contour lines, and the contours created by billboards attached to the exterior walls of the buildings as secondary contour lines, arguing that it is the primary contour lines that contribute to the beauty of the streetscape, therefore we should try to reduce the secondary contours as much as possible. When Ashihara began his research in the 1970s, digital technology was still undeveloped, so he analyzed the streets of Ginza using his hands and eyes. We tried to automate Ashihara's ideas with digital technology. Our goal was to extract architectural outlines (primary contour lines) and billboards (secondary contour lines) from the large volume of landscape images we had collcted. However, most of the algorithms for detecting objects in landscape images were created for Western cityscapes, and there were not many suitable for detecting outdoor advertisements or billboards, which are unique to Asian landscapes including Japan, so we started by creating an algorithm. Specifically, we had to perform the steady task of finding advertisements in the landscape images with the human eye, and then paint these areas with color by hand. In order to create a machine-learning model with sufficient accuracy, tens of thousands of such colored images would be required. Beyond that necessity, however, there is the possibility of providing an objective framework for ambiguous sensations such as "beauty," which are often left to the subjective views of the individual. Through evaluations obtained through the eyes of a machine and its processes, this project demonstrates the possibility of shedding light from a different perspective on the meaning of beauty, and the meaning of the urban landscape for us humans.
展开▼
