TY  - JOUR
AU  - Dang, Zifeng
AU  - Liu, Jiahan
AU  - Xiong, Shuang
AU  - Ma, Jun
AU  - Ren, Xunhuan
AU  - Tsviatkou, Viktar Yurevich
PY  - 2026
DA  - 2026/03/29
TI  - AMRC-NET: A New Method for Recognition of Russian Handwritten Text Integrating Multipath Mechanism and Linguistic Features
JO  - Chinese Journal of Information Fusion
T2  - Chinese Journal of Information Fusion
JF  - Chinese Journal of Information Fusion
VL  - 3
IS  - 1
SP  - 62
EP  - 73
DO  - 10.62762/CJIF.2025.868838
UR  - https://www.icck.org/article/abs/CJIF.2025.868838
KW  - Russian handwritten text recognition
KW  - AMRC-NET
KW  - attention mechanism
KW  - multi-scale feature fusion
KW  - adaptive model integration
KW  - optical character recognition
AB  - Russian handwritten text recognition presents significant challenges due to the complex morphology of the Cyrillic alphabet, prevalent cursive writing, and substantial writer variability. To address the limitations of existing methods in dynamic contextual modeling and language-specific feature adaptation, this paper proposes an end-to-end framework named AMRC-NET. This framework integrates a multi-path architecture with linguistic feature awareness through three core modules: a Context Enhancement Module for long-range dependency modeling, a Russian Alphabet Morphology Optimization Module for script-specific pattern capture, and a Multi-Path Adaptive Fusion Mechanism for dynamic output integration. Extensive experiments on the Cyrillic Handwriting Dataset demonstrate that AMRC-NET achieves 48.15% in recognition accuracy, significantly outperforming existing models such as STAR-Net and GRCNN. Ablation studies further reveal that the three modules contribute accuracy improvements of 4.82%, 2.03%, and 1.58%, respectively, confirming their individual effectiveness and synergistic integration. The model also exhibits enhanced generalization capability, as reflected by lower validation loss and a reduced overfitting gap, while visual analysis confirms its robustness in challenging cursive writing scenarios.
SN  - 2998-3371
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Dang2026AMRCNET,
  author = {Zifeng Dang and Jiahan Liu and Shuang Xiong and Jun Ma and Xunhuan Ren and Viktar Yurevich Tsviatkou},
  title = {AMRC-NET: A New Method for Recognition of Russian Handwritten Text Integrating Multipath Mechanism and Linguistic Features},
  journal = {Chinese Journal of Information Fusion},
  year = {2026},
  volume = {3},
  number = {1},
  pages = {62-73},
  doi = {10.62762/CJIF.2025.868838},
  url = {https://www.icck.org/article/abs/CJIF.2025.868838},
  abstract = {Russian handwritten text recognition presents significant challenges due to the complex morphology of the Cyrillic alphabet, prevalent cursive writing, and substantial writer variability. To address the limitations of existing methods in dynamic contextual modeling and language-specific feature adaptation, this paper proposes an end-to-end framework named AMRC-NET. This framework integrates a multi-path architecture with linguistic feature awareness through three core modules: a Context Enhancement Module for long-range dependency modeling, a Russian Alphabet Morphology Optimization Module for script-specific pattern capture, and a Multi-Path Adaptive Fusion Mechanism for dynamic output integration. Extensive experiments on the Cyrillic Handwriting Dataset demonstrate that AMRC-NET achieves 48.15\% in recognition accuracy, significantly outperforming existing models such as STAR-Net and GRCNN. Ablation studies further reveal that the three modules contribute accuracy improvements of 4.82\%, 2.03\%, and 1.58\%, respectively, confirming their individual effectiveness and synergistic integration. The model also exhibits enhanced generalization capability, as reflected by lower validation loss and a reduced overfitting gap, while visual analysis confirms its robustness in challenging cursive writing scenarios.},
  keywords = {Russian handwritten text recognition, AMRC-NET, attention mechanism, multi-scale feature fusion, adaptive model integration, optical character recognition},
  issn = {2998-3371},
  publisher = {Institute of Central Computation and Knowledge}
}