开源两款新型虹膜识别模型，降低参与国家级评测门槛。
Lowering the Barrier to IREX Participation: Open-Source Algorithms, Toolkit, and Benchmarking for Iris Recognition
- 提出两个基于深度学习的开源虹膜匹配算法，适配IREX标准
- 首次实现开源虹膜算法进入IREX X排行榜，性能达可比肩商用系统
- 提供完整C++/Python实现与评测工具，适合研究者快速验证
NIST虹膜交换（IREX）为评估新型开源虹膜识别算法提供了优质平台，但其高准入门槛——要求使用C++、特定API，并满足严格的运行速度与内存限制——阻碍了广泛参与。本文旨在降低这些门槛，推动开源虹膜识别的大规模评测：(a) 提出两个现代深度学习架构的开源虹膜匹配器ArcIris与TripletIris，及其符合IREX X规范的C++实现，首次被纳入IREX X排行榜；同时发布可集成的新分割与虹膜圆度逼近模型；(b) 按IREX X测试协议对所有主流开源虹膜识别方案进行性能评估。论文还提供ArcIris与TripletIris的Python实现，分析C++与Python实现间的差异。此外，公开了两个现有方法的IREX X兼容版C++实现：(a) 基于人眼注意驱动核的虹膜图像滤波算法（HDBIF），(b) 可解释的弗克斯斑点检测与比对算法（CRYPTS）。除报告IREX X结果外，还在多个学术基准上评估了所有方法性能：Q-FIRE、Warsaw-Biobase Post-Mortem Iris、CASIA-Iris-Thousand-V4、CASIA-Iris-Lamp-V4、IIT Delhi Iris Database、IIITD Contact Lens Iris Database、NDIris3D、Notre Dame Variable Iris Image Quality Release 2 (VII-Q-R2)。
原文摘要 · Abstract (English)
NIST Iris Exchange (IREX) offers an appealing solution to evaluating new open-source iris recognition algorithms, but it presents high barriers to entry because these algorithms must be written in C++, using a specific API, and adapted to meet strict IREX speed and memory constraints. The main goal of this paper is to lower these barriers and advance open-source iris recognition large-scale evaluations by offering: (a) two new modern deep learning-based open-source iris matchers (ArcIris and TripletIris), along with their C++ IREX X-compliant implementations, which are the first open-source iris recognition methods included into the IREX X leaderboard (and thus IREX-vetted), as well as new segmentation and iris circular approximation models that can be incorporated into any new iris recognition method, and (b) a performance assessment (according to IREX X testing protocols) of all major and currently available open-source iris recognition solutions. The paper also provides Python implementations of the new ArcIris and TripletIris methods and discusses the differences one may encounter between C++ and Python implementations of the same conceptually equivalent approaches. Finally, the paper offers open-source, IREX X-compliant C++ implementations of two existing methods: (a) an iris image filtering-based algorithm utilizing human saliency-driven kernels (HDBIF), and (b) a human-interpretable algorithm for detecting and comparing Fuchs' crypts (CRYPTS). In addition to IREX X evaluation results, the paper reports the performance of all methods on major academic benchmarks: Quality-Face/Iris Research Ensemble (Q-FIRE), Warsaw-Biobase Post-Mortem Iris, CASIA-Iris-Thousand-V4, CASIA-Iris-Lamp-V4, IIT Delhi Iris Database, IIITD Contact Lens Iris Database, NDIris3D, and Notre Dame Variable Iris Image Quality Release 2 (VII-Q-R2).
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。