SecMeanshift: FSS-based privacy-preserving mean-shift clustering

Clustering is an unsupervised learning method that groups data based on similar characteristics. They have broad applications in various fields, such as image processing, text clustering, and recommendation systems. However, in practical scenarios, clustering often involves sensitive data from multiple data owners, which raises significant privacy concerns. Therefore, addressing the challenge of performing efficient clustering while preserving data privacy is critical. Existing privacy-preserving clustering methods often encounter challenges such as high computational overhead or reliance on auxiliary information. Consequently, we propose SecMeanshift, a privacy-preserving mean-shift framework based on function secret sharing. SecMeanshift leverages an offline–online paradigm to enhance efficiency by offloading some operations to the offline phase. We design basic protocols, including secure negative exponential, secure select, and secure sampling for the proposed framework, and design private protocols for each phase of the mean-shift. Theoretical analysis confirms the security and correctness of the proposed protocols. Furthermore, extensive experiments on diverse datasets demonstrate that SecMeanshift achieves significantly higher efficiency than HE-Meanshift, making it a promising solution for privacy-preserving clustering.