Bézier Curve-Based Stroke Extraction for Handwritten Characters


Yuxuan TENG (Tokyo University of Science), Takuya Matsuzaki (Tokyo University of Science)
The 35th British Machine Vision Conference

Abstract

Accurate stroke extraction from handwritten character images is essential for realistic handwriting generation and stroke-based character recognition. Traditional pixel-based segmentation methods often produce fragmented strokes, failing to preserve accurate topological structures. To solve this, we propose a novel Bézier curve-based stroke extraction method. Our approach employs deep learning-based image registration to initially align handwritten images with vector-based reference fonts, followed by precise Bézier curve fitting to the skeleton of the character image. Evaluations on a Japanese handwritten character dataset show that characters reconstructed from extracted strokes achieve high OCR recognition accuracy (95.58%), indicating accurate representation of the original handwriting structures. Additionally, geometric similarity evaluated via feature-point matching confirms precise stroke extraction, though accuracy slightly decreases for complex characters with many strokes. Moreover, our Bézier-based representation naturally supports interpolation between character images by linearly interpolating corresponding control points, enabling the synthesis of diverse handwritten samples. Such synthesized variations hold promise for augmenting OCR training datasets as well as providing a succinct representation of a sample space for stroke-based character image generation.

Citation

@inproceedings{TENG_2025_BMVC,
author    = {Yuxuan TENG and Takuya Matsuzaki},
title     = {Bézier Curve-Based Stroke Extraction for Handwritten Characters},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_478/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection