Positional Component-Guided Hangul Font Image Generation via Deep Semantic Segmentation and Adversarial Style Transfer

This work proposes a YOLOv8-driven, positional-aware decomposition of Hangul characters into initial, medial, and final components, which are then recombined via a GAN to synthesize fonts that better preserve structure and style than existing methods (MXFont, CKFont). The framework achieves lower L1/FID and higher SSIM metrics, enabling finely controllable and scalable multilingual typography generation.

Recommended citation: Abdul Sami, Jaeyong Choi. (2025). "Positional Component-Guided Hangul Font Image Generation via Deep Semantic Segmentation and Adversarial Style Transfer." MDPI Electronics. 14(13).
Download Paper