Scholarlink

With the rapid development of biometrics, palmprint recognition, due to uniqueness, non-contact and high security, is a core intelligent authentication technology. However, cross-device domain shift from hardware differences and complex environments degrades recognition accuracy and stability. To address this, we propose TSCAN, a teacher-student co-learning network for cross- device palmprint recognition. It introduces adaface loss to adjust classification boundaries, tackling poor discriminability in low-quality images. A teacher- student framework with pseudo-label transmission and EMA updates enables learning unlabeled target domain data. Adversarial learning combined with GRL aligns source and target feature distributions. Experiments on XJTU-UP show that TSCAN improves recognition accuracy by 7.5% in four cross-device scenarios and reduces EER by 8.5% compared to the baseline, and excellent generalization across target domains, validating its practical value in intelligent authentication.

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

TSCAN: Teacher-Student Co-Learning Adaptive Network for Cross-Device Palmprint Recognition

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Huayang Li,
  • Huikai Shao,
  • Yani Ren,
  • Dexing Zhong

摘要

With the rapid development of biometrics, palmprint recognition, due to uniqueness, non-contact and high security, is a core intelligent authentication technology. However, cross-device domain shift from hardware differences and complex environments degrades recognition accuracy and stability. To address this, we propose TSCAN, a teacher-student co-learning network for cross- device palmprint recognition. It introduces adaface loss to adjust classification boundaries, tackling poor discriminability in low-quality images. A teacher- student framework with pseudo-label transmission and EMA updates enables learning unlabeled target domain data. Adversarial learning combined with GRL aligns source and target feature distributions. Experiments on XJTU-UP show that TSCAN improves recognition accuracy by 7.5% in four cross-device scenarios and reduces EER by 8.5% compared to the baseline, and excellent generalization across target domains, validating its practical value in intelligent authentication.