U-Net Handwriting Removal Method and Dataset with ResNetV2 Fusion
Journal of Image Processing Theory and Applications · 2026 · 9 (1)
DOI: 10.23977/jipta.2026.090102 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Image Processing Theory and Applications, 2026.
Abstract (Originalsprache)
: With the growing demand for paperless offices and digital archiving, many paper documents are scanned into image formats for management and dissemination. These images often contain handwritten annotations, signatures, or markings, which interfere with accurate understanding and automatic analysis, especially in educational scenarios where exams and assignments include extensive handwritten content. This highlights the need for effective handwritten text removal techniques. This work proposes an end-to-end handwritten text removal method based on a U-Net enhanced with ResNetV2 modules. The model leverages multi-scale feature extraction, residual learning, and skip connections to remove handwritten marks while preserving printed text and document layout. In addition, a high-quality, large-scale handwritten text removal dataset is constructed and publicly released to provide a standardized benchmark for evaluation and reproducibility. Experimental results show that the proposed approach efficiently removes handwritten traces while maintaining document structure and visual consistency, improving the usability of digital documents. The study contributes to research on handwritten text removal and provides technical support for educational resource digitization, smart learning, and document management.
Zitieren
APA
(2026). U-Net Handwriting Removal Method and Dataset with ResNetV2 Fusion. Journal of Image Processing Theory and Applications, 9(1). https://doi.org/10.23977/jipta.2026.090102
Vancouver
U-Net Handwriting Removal Method and Dataset with ResNetV2 Fusion. Journal of Image Processing Theory and Applications. 2026;9(1). doi:10.23977/jipta.2026.090102
BibTeX
@article{euvimed2026,
title = {U-Net Handwriting Removal Method and Dataset with ResNetV2 Fusion},
journal = {Journal of Image Processing Theory and Applications},
year = {2026},
volume = {9},
number = {1},
doi = {10.23977/jipta.2026.090102}
}
EUVIMED