Secure Approximate Nearest Neighbor Search over Encrypted Data

Abstract

For the past decade, approximate nearest neighbor (ANN) search in high dimensional space has been studied extensively. However, it supports only ANN search over palintext in traditional locality sensitive hashing (LSH). How to perform ANN search over encrypted data becomes a new challenging task. In this paper, we make an attempt to formally address the problem. We propose a new secure and efficient ANN search scheme over encrypted data based on Sorting Keys-LSH (LSH) and mutable order-preserving encryption (mOPE). In our construction, we exploit SK-LSH to generate indexes locally. While data and indexes should be outsourced to the cloud in encrypted form, which complicates computations on the encrypted data. Then, we encrypt LSH indexes using mOPE for efficient ANN search. Through rigorous security and efficiency analysis, we show that our proposed scheme is secure under the proposed model, while correctly realizing the goal of secure ANN search over encrypted data.