Scholarlink

With the widespread application of facial recognition technology, liveness detection has become a critical component in ensuring system security. Unlike interactive liveness detection, silent liveness detection has significant advantages in practical applications, as it does not require user cooperation. However, its generalization ability in cross-domain scenarios remains a substantial challenge. To address this, this paper proposes a silent liveness detection method based on adversarial domain learning, which enhances the model’s adaptability to unseen attacks through a domain feature expansion mechanism. This approach introduces adversarial learning strategies to generate representations with domain-shifted features, forcing the model to learn more robust, domain-invariant discriminative features. As a result, the method significantly improves detection robustness against previously unseen attack types. Extensive cross-domain experimental results demonstrate that the proposed method achieves outstanding performance across multiple benchmark datasets, considerably outperforming existing methods in attack detection accuracy. This study provides an effective solution to the domain generalization problem in silent liveness.

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

Silent Face Liveness Detection Enhanced by Adversarial Domain Learning for Domain Feature Expansion

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jiali Gong,
  • Ying Tang,
  • Jingyu Liu,
  • Senting Wang,
  • Chunyan Fu,
  • Zhaojuan Zhang,
  • Xiaoqing Wu,
  • Mengke Xu,
  • Wanli Huo

摘要

With the widespread application of facial recognition technology, liveness detection has become a critical component in ensuring system security. Unlike interactive liveness detection, silent liveness detection has significant advantages in practical applications, as it does not require user cooperation. However, its generalization ability in cross-domain scenarios remains a substantial challenge. To address this, this paper proposes a silent liveness detection method based on adversarial domain learning, which enhances the model’s adaptability to unseen attacks through a domain feature expansion mechanism. This approach introduces adversarial learning strategies to generate representations with domain-shifted features, forcing the model to learn more robust, domain-invariant discriminative features. As a result, the method significantly improves detection robustness against previously unseen attack types. Extensive cross-domain experimental results demonstrate that the proposed method achieves outstanding performance across multiple benchmark datasets, considerably outperforming existing methods in attack detection accuracy. This study provides an effective solution to the domain generalization problem in silent liveness.