In this paper, we describe a novel Privacy Preserving Biometric Authentication (PPBA) system designed for Mobile Edge Computing (MEC) and multimodal biometrics. We focus on hill climbing attacks that reveal biometric templates to insider adversaries despite the encrypted storage in the cloud. First, we present an impossibility result on the existence of two-party PPBA systems that are resistant to these attacks. To overcome this negative result, we add a non-colluding edge server for detecting hill climbing attacks both in semi-honest and malicious model. The edge server that stores each user’s secret parameters enables to outsource the biometric database to the cloud and perform matching in the encrypted domain. The proposed system combines Set Overlap and Euclidean Distance metrics using score level fusion. Here, both the cloud and edge servers cannot learn the fused matching score. Moreover, the edge server is prevented from accessing any partial score. The efficiency of the crypto-primitives employed for each biometric modality results in linear computation and communication overhead. Under different MEC scenarios, the new system is found to be most efficient with a 2-tier architecture, which achieves %75 lower latency compared to mobile cloud computing.

在本文中，我们描述了一种为移动边缘计算(MEC) 和多模态生物识别设计的新型隐私保护生物识别认证 (PPBA) 系统。我们专注于爬山攻击，尽管在云中进行了加密存储，但这些攻击会向内部对手揭示生物识别模板。首先，我们提出了一个不可能的结果，即存在能够抵抗这些攻击的两方 PPBA 系统。为了克服这个负面结果，我们添加了一个非共谋边缘服务器，用于在半诚实和恶意模型中检测爬山攻击。边缘服务器存储每个用户的秘密参数，可以将生物特征数据库外包到云端并在加密域中进行匹配。提议的系统结合了 Set Overlap 和使用分数级融合的欧几里德距离度量。在这里，云服务器和边缘服务器都无法学习融合匹配分数。此外，防止边缘服务器访问任何部分分数。用于每种生物特征模式的加密原语的效率导致线性计算和通信开销。在不同的 MEC 场景下，发现新系统采用 2 层架构的效率最高，与移动云计算相比，延迟降低了 75%。