Scholarlink

<p>Aiming at the problems of poor interpretability and low cross-scenario structural generality in traditional iris recognition, this paper proposes an iris recognition framework based on statistical clustering. Combining classical computer vision with statistical learning, the framework processes iris images through feature transformation, normalization, and pooling. Subsequently, it determines a 32-dimensional continuous data range as category empirical knowledge via cluster analysis, and realizes one-to-many recognition and unlimited addition of new categories based on multi-parameter collaborative similarity calculation. Experimental results show that on the JLU and CASIA iris databases, the framework achieves an average Equal Error Rate (EER) as low as 0.72%-0.78%, with a single-image inference time of only 1.5 ms in a CPU environment. Moreover, the addition of new categories does not interfere with the performance of the original model. This framework balances recognition accuracy, structural generality, and operational interpretability, while featuring low computational cost, making it suitable for resource-constrained scenarios.</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

Statistical clustering-based iris recognition: Addressing interpretability and uncertainty challenges

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Shuai Liu

摘要

Aiming at the problems of poor interpretability and low cross-scenario structural generality in traditional iris recognition, this paper proposes an iris recognition framework based on statistical clustering. Combining classical computer vision with statistical learning, the framework processes iris images through feature transformation, normalization, and pooling. Subsequently, it determines a 32-dimensional continuous data range as category empirical knowledge via cluster analysis, and realizes one-to-many recognition and unlimited addition of new categories based on multi-parameter collaborative similarity calculation. Experimental results show that on the JLU and CASIA iris databases, the framework achieves an average Equal Error Rate (EER) as low as 0.72%-0.78%, with a single-image inference time of only 1.5 ms in a CPU environment. Moreover, the addition of new categories does not interfere with the performance of the original model. This framework balances recognition accuracy, structural generality, and operational interpretability, while featuring low computational cost, making it suitable for resource-constrained scenarios.