Fuzzy Private Hash Matching for Harmful Media Moderation in End-to-End Encrypted Communication

End-to-End Encryption (E2EE) effectively prevents messaging platforms from accessing media content but limits the automatic detection of harmful media using popular perceptual hash-matching methods. Recently, Private Hash Matching (PHM) techniques have shown great potential for moderating harmful content in E2EE. However, many state-of-the-art PHM schemes either focus solely on exact hash matching, neglecting the more practical fuzzy hash matching, or fail to balance privacy, efficiency, and robustness. In this paper, we present an efficient and robust Fuzzy Private Hash Matching (FPHM) scheme for harmful media moderation in E2EE. Our scheme comprises two steps: 1) coarse matching on the client side that filters out the legal media using an index structure based on a Dynamic-Encoding-Tree Locality-Sensitive Hash (DET-LSH), 2) fine-grained, distance-aware fuzzy matching through client-server interaction on the filtered data. Security analyses confirm that the scheme robustly protects both user media information and the server's harmful hash set from unauthorized disclosure. The experiment indicates that FPHM enhances inspection efficiency and robustness with negligible false negatives.