| Title |
Improvement of a Segmentation Network for Character Stroke Extraction from Metal Movable Type Printed Documents |
| Authors |
이우석(Woo-Seok Lee) ; 최강선(Kang-Sun Choi) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.12.31 |
| Keywords |
Stroke segmentation; Loss function; Historical printed documents; Metal movable type |
| Abstract |
In this paper, we analyze loss functions for the stroke extraction process from old documents printed with movable metal types, and propose a new loss function to improve its performance. The proposed loss function, boundary gradient consistency loss, makes the gradient of the stroke boundary consistent to produce smooth stroke boundary. The stroke extraction network trained with the proposed loss achieves the highest score in terms of the Dice coefficient and smoothness. And it achieves similar score to gradient difference loss that achieves the highest score in terms of noise suppression. |