Scholarlink

Iris images inpainting is an effective means to reduce the impact of low-quality iris images in recognition. To improve the recognition rate of low-quality iris image, this paper proposes a High-Low Feature Fusion Generative Adversarial Network (HLFG) to restore irregularly occluded iris images. HLFG designs hierarchical attention to capture long-distance dependencies and enhance context delivery, and designs feature fusion up-sampling modules in skip connections to fuse local details and global structure. The experimental results based on three iris datasets, CASIA-Iris-Interval, IITD and ND-IRIS-0405, show that the PNSR of repaired iris images increased by 20.2521dB, 19.4234dB and 26.0561dB respectively; SSIM increased by 0.0897, 0.1090 and 0.1542, respectively; TAR increased by 27.13%, 8.94% and 53.55%, respectively; EER decreased by 4.6337%, 1.8876% and 12.8380%, respectively and the above results fully validate the effectiveness of the HLFG.

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

High-Low Feature Fusion Generative Adversarial Network for the Inpainting of Irregularly Occluded Iris Images

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Junkang Deng,
  • Zhijie Chen,
  • Ying Chen,
  • Changle He,
  • Xiaodong Zhu

摘要

Iris images inpainting is an effective means to reduce the impact of low-quality iris images in recognition. To improve the recognition rate of low-quality iris image, this paper proposes a High-Low Feature Fusion Generative Adversarial Network (HLFG) to restore irregularly occluded iris images. HLFG designs hierarchical attention to capture long-distance dependencies and enhance context delivery, and designs feature fusion up-sampling modules in skip connections to fuse local details and global structure. The experimental results based on three iris datasets, CASIA-Iris-Interval, IITD and ND-IRIS-0405, show that the PNSR of repaired iris images increased by 20.2521dB, 19.4234dB and 26.0561dB respectively; SSIM increased by 0.0897, 0.1090 and 0.1542, respectively; TAR increased by 27.13%, 8.94% and 53.55%, respectively; EER decreased by 4.6337%, 1.8876% and 12.8380%, respectively and the above results fully validate the effectiveness of the HLFG.