| Title |
Quantification and Analysis Methods of Architectural Styles Using Image-Text AI Models |
| Authors |
유영진(Yoo, Youngjin) ; 이진국(Lee, Jin-Kook) |
| DOI |
https://doi.org/10.5659/JAIK.2025.41.5.93 |
| Keywords |
Architectural Style Quantification; Architectural Style Analysis; Image-Text AI; Style Features; Cluster Analysis |
| Abstract |
This study introduces a quantitative method for analyzing architectural styles, addressing the limitations of traditional qualitative approaches.
An objective measurement system was developed using an Image-Text AI model to analyze architectural images and convert visual
characteristics into numerical values. The analysis focuses on four key features: curvature, color saturation, transparency, and symmetry. Each
feature is measured on a normalized scale between two opposing states, with the results including mean values, standard deviations, and
standout characteristics. The method was validated using approximately 9,000 images from works by 30 renowned architects, including
Pritzker Prize winners. Feature modification tests confirmed consistent results, with minimal variation (±0.05) in features that were not altered.
Clustering analysis revealed meaningful patterns within individual architects’ work and across different architects, aligning closely with
traditional qualitative classifications. This method provides a structured framework for objectively comparing and classifying architectural
styles, especially valuable as digital visualization becomes more common in architectural design. Although currently limited to four visual
features, the approach lays a foundation for future expansion, with potential applications in style evaluation, building similarity searches,
architectural database organization, and curated style exploration. |