HF-Font: Few-Shot Font Generation via High-Frequency Style Enhancement and Fusion

Weiran Chen, Guiqian Zhu, Ying Li, Yi Ji, Chunping Liu

Keywords: Style Transfer; Image-to-Image Translation; Font Generation; Deep Generative Model

TL;DR: In this paper, we propose HF-Font, a novel framework that generates fonts with higher structural fidelity and quality.

Abstract: Few-shot font generation aims to create new fonts with a limited number of glyph references. It can be used to greatly reduce the workload of manual font design. However, although existing methods have achieved satisfactory performance, they still struggle to capture delicate glyph details, thus resulting in stroke errors, artifacts, and blurriness. To address these problems, we propose HF-Font, a novel framework that generates fonts with higher structural fidelity. Specifically, inspired by the observation that high-frequency information of character images often contains distinct style patterns (e.g., glyph topology and stroke variation), we develop a novel style-enhanced module to improve the style extraction by incorporating high-frequency features from reference images using a high-pass filter. Then, for guiding the generation process, we design a Style-Content Fusion Module (SCFM), which integrates the style features with a component-wise codebook that encodes content semantics. Moreover, we also introduce a style contrastive loss to better transfer high-frequency features. Extensive experiments show that our HF-Font outperforms the state-of-the-art methods in both qualitative and quantitative evaluations, demonstrating its effectiveness across diverse font styles and characters. Our source code will be released soon.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 6712