Korean Journal of Computational Design and Engineering

This study proposes a quantification methodology for architectural design styles and develops evaluation systems utilizing this approach. The research establishes a quantification framework based on CLIP multimodal AI that objectively measures key visual characteristics such as curvature, saturation, transparency, and symmetry. Two evaluation systems were developed: a multiple criteria-based classification system and a single criterion-based curation system. The systems were integrated with an AI visualization tool to create a unified workflow from image generation to evaluation. Verification tests using works of 20 Pritzker Prize-winning architects showed classification accuracies of 89% for actual architectural photographs and 81-82% for AI-generated images. The developed systems demonstrated their effectiveness in automated style classification and curation, offering practical tools for architectural design evaluation and management in the generative AI era.

키워드 Architectural style quantification, Design evaluation, Multimodal AI, Style classification, Design curation