Scholarlink

<p>The verification of passport authenticity is crucial for global security and border management. The classification of passport background texture, as a vital anti-counterfeiting measure, plays a pivotal role in this process. However, existing methods often struggle with achieving high recognition accuracy. In this study, we propose TCNet, a complementary framework that integrates Convolutional Neural Networks (CNNs) and Transformers. TCNet leverages the local feature extraction capabilities of CNNs and the global dependency modeling of Transformers to enhance the performance of passport background texture classification. Furthermore, we introduce a Dynamic Weighting Module (DWM) to achieve adaptive weighted fusion of these two feature types, thereby strengthening the representation of background texture features. Experimental results demonstrate that TCNet outperforms other mainstream models on the passport background texture dataset, achieving an accuracy of 86.40%. Here, we show that TCNet holds significant potential for application in passport authenticity detection, providing a robust solution that balances efficiency and accuracy. The code and dataset are available at <a href="https://github.com/braverSheep/TCNet">https://github.com/braverSheep/TCNet</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid CNN-transformer framework with dynamic feature fusion for enhanced passport background texture classification

  • 期刊论文

  • 发表日期：

    2025年12月2日

查看全文
  • Maoqin Tian,
  • Lin Tang,
  • Jiafeng Xu,
  • Yibo Zhang,
  • Yong Yang,
  • Lingpei Zeng,
  • Eryang Chen,
  • Yuanlun Xie

摘要

The verification of passport authenticity is crucial for global security and border management. The classification of passport background texture, as a vital anti-counterfeiting measure, plays a pivotal role in this process. However, existing methods often struggle with achieving high recognition accuracy. In this study, we propose TCNet, a complementary framework that integrates Convolutional Neural Networks (CNNs) and Transformers. TCNet leverages the local feature extraction capabilities of CNNs and the global dependency modeling of Transformers to enhance the performance of passport background texture classification. Furthermore, we introduce a Dynamic Weighting Module (DWM) to achieve adaptive weighted fusion of these two feature types, thereby strengthening the representation of background texture features. Experimental results demonstrate that TCNet outperforms other mainstream models on the passport background texture dataset, achieving an accuracy of 86.40%. Here, we show that TCNet holds significant potential for application in passport authenticity detection, providing a robust solution that balances efficiency and accuracy. The code and dataset are available at https://github.com/braverSheep/TCNet.