Scholarlink

The fingerprint identification system is widely used for personnel identity confirmation. However, fingerprint contamination, such as sweat, reduces accuracy. We proposed MLDN-Net, which employs a recurrent neural network with a KNN classifier and CoordConv layers, enabling adaptation to wet fingerprints. Tested on the Nasic9395 1018 dataset, MLDN-Net reduced FRR from 19.6% to 9.2%, outperforming state-of-the-art denoising models.

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

MLDN-Net: Multi-Level Denoising Neural Network for Wet Fingerprint via Cyclic Multi-Variate Function

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chao-Lin Kuo,
  • Ching-Te Chiu,
  • Mao-Hsiu Hsu

摘要

The fingerprint identification system is widely used for personnel identity confirmation. However, fingerprint contamination, such as sweat, reduces accuracy. We proposed MLDN-Net, which employs a recurrent neural network with a KNN classifier and CoordConv layers, enabling adaptation to wet fingerprints. Tested on the Nasic9395 1018 dataset, MLDN-Net reduced FRR from 19.6% to 9.2%, outperforming state-of-the-art denoising models.