Abstract With the escalated usage of a biometric authentication system (BAS), template protection for biometrics attracted research interest in recent years. The assumption behind the existing homomorphic encryption-based BASs is that the server performs the computations honestly. In a malicious server setting, the server may return an arbitrary result to save the computational resources, which may result in false accept/reject. To tackle this challenge, we propose a secure and verifiable classification based iris authentication system (SvaS). SvaS aims to achieve both privacy-preserving (PP) training and PP classification of Nearest Neighbor and Multi-class Perceptron models. The Fan-vercauteren scheme provides confidentiality for the iris templates, and aggregate verification vector helps to verify the correctness of the computed classification result. Extensive experimental results on benchmark iris databases demonstrate that SvaS provides privacy to the iris templates with no loss in accuracy and eliminates the need to trust the server.

中文翻译：

---

使用全同态加密的安全且可验证的虹膜认证系统

摘要 随着生物特征认证系统（BAS）的使用升级，生物特征模板保护近年来引起了研究兴趣。现有的基于同态加密的 BAS 背后的假设是服务器诚实地执行计算。在恶意服务器设置中，服务器可能会返回任意结果以节省计算资源，这可能会导致错误的接受/拒绝。为了应对这一挑战，我们提出了一种基于安全和可验证分类的虹膜认证系统（SvaS）。SvaS 旨在实现最近邻和多类感知器模型的隐私保护 (PP) 训练和 PP 分类。Fan-vercauteren 方案为虹膜模板提供机密性，聚合验证向量有助于验证计算出的分类结果的正确性。基准虹膜数据库上的大量实验结果表明，SvaS 为虹膜模板提供了隐私，而不会损失准确性，并且无需信任服务器。