Multi-view Handwritten Character Recognition Method with Adversarial Domain Adaptation and Momentum Prototype Fusion.

Saved in:

Bibliographic Details
Title:	Multi-view Handwritten Character Recognition Method with Adversarial Domain Adaptation and Momentum Prototype Fusion.
Authors:	Zhang, Moyi¹ 232085404022@lut.edu.cn, Zhang, Bin² 17344063618@163.com, Pei, Shaofeng² 1143945122@qq.com, Cheng, Haiyan¹ 19029334977@163.com
Source:	IAENG International Journal of Computer Science. Jun2026, Vol. 53 Issue 6, p2099-2112. 14p.
Subjects:	Feature extraction, Pattern recognition systems
Abstract:	Multi-view handwritten character recognition faces data distribution shifts due to viewpoint and environmental changes. To address this, we propose a synergistic framework composed of three core modules: a structure-enhanced feature extractor, an adversarial training module, and a momentum prototype learning component. Together, they enable effective adversarial domain adaptation. First, the feature extractor fuses Channel Attention and Deformable Convolution to capture complex geometric deformations. Next, adversarial training reduces the distribution gap between domains, facilitating the learning of domain-invariant representations. Finally, a momentum prototype module aligns intra-class structures using dynamically updated prototypes, enhancing feature separability. These modules are jointly optimized to improve discriminative power. Experimental results demonstrate superior performance on multiple cross-domain datasets. The method achieves an average accuracy of 95.7% on synthetic multi-view handwritten datasets and an average accuracy of 83.9% on the Office-31 benchmark. Notably, on a challenging real-world multi-view digit dataset, it reaches 93.6% accuracy. These results confirm the model's robustness in bridging the simulation-to-reality gap. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Multi-view handwritten character recognition faces data distribution shifts due to viewpoint and environmental changes. To address this, we propose a synergistic framework composed of three core modules: a structure-enhanced feature extractor, an adversarial training module, and a momentum prototype learning component. Together, they enable effective adversarial domain adaptation. First, the feature extractor fuses Channel Attention and Deformable Convolution to capture complex geometric deformations. Next, adversarial training reduces the distribution gap between domains, facilitating the learning of domain-invariant representations. Finally, a momentum prototype module aligns intra-class structures using dynamically updated prototypes, enhancing feature separability. These modules are jointly optimized to improve discriminative power. Experimental results demonstrate superior performance on multiple cross-domain datasets. The method achieves an average accuracy of 95.7% on synthetic multi-view handwritten datasets and an average accuracy of 83.9% on the Office-31 benchmark. Notably, on a challenging real-world multi-view digit dataset, it reaches 93.6% accuracy. These results confirm the model's robustness in bridging the simulation-to-reality gap. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X