ScholarWorks@숙명여자대학교

초록

This study presents a cross-script font recommendation system that preserves visual consistency between Korean and Chinese fonts. The proposed framework adopts a two-stage approach. First, K-Means clustering is applied to build inter-cluster mappings that narrow the candidate space. Second, a three-branch encoder (style, structure, stroke) with attention-based fusion generates multi-feature embeddings, and multi-dimensional similarity measures produce Top-N recommendations. Experiments with 1,600 fonts demonstrated the feasibility of this framework through case studies, showing that it can capture style, structure, and stroke similarities in an interpretable way. A usability study with 15 participants, including Chinese speakers and design/engineering students, further confirmed its practical value. Over 96% of participants agreed that visually consistent fonts appeared in the Top-5 results, and all evaluation items scored above 4.0 on a 5-point scale. These results indicate that the two-stage design combining cluster mapping and attention-based multi-feature embeddings improves efficiency and provides perceptually convincing outcomes. The system offers a practical tool for multilingual publishing, UI/UX design, and brand identity management, with potential for extension to broader datasets and diverse writing systems.

키워드