Nearest neighbour search over encrypted data using intel SGX

Abstract Content-based image retrieval (CBIR) retrieves desired digital images from large databases. “Content-based” means that the search analyzes the contents of the image rather than the metadata such as keywords, tags, or descriptions associated with images. It has different applications in different domains such as crime prevention, intellectual property, medical diagnosis, and face finding. Based on the aforementioned applications, there is a desideratum to retrieve users who have content-wise similar images, e.g., patients to speedup the diagnosis process or individuals to identify an unidentified criminals or group individuals with similar interests. Most image owners are outsourcing their images to the cloud because of low storage and computation costs. Although existing encryption mechanisms protect images from unauthorized access, these techniques increase the computational complexity of executing arbitrary functions on the outsourced images. In this paper, our focus is to efficiently and privately identify users who have content-wise similar encrypted images stored in the cloud. The proposed scheme utilizes the Intel Software Guard Extensions (Intel SGX) architecture, the Convolutional Neural Network (CNN), and Locality Sensitive Hashing (LSH) for minhash signatures. Considering symmetric encryption, we experimentally show that the proposed approach is only five times slower than the plaintext approach with just one round of interaction with the cloud server.