Privacy-preserving content-based image retrieval for mobile computing

Abstract

With the increasing popularity of mobile cloud computing, image outsourcing provides a cost-effective way to support images management for smart mobile devices. However, users default that the cloud server is not trustworthy, secure image retrieval has attracted considerable interests recently to address users’ privacy concerns. Due to the limited resources such as computing power, battery lifetime, and storage capacity, it still suffers from the challenges of relieving mobile devices of excessive computation burdens, such as feature extraction and index building. In this paper, we propose a framework that supports cloud server side local-feature (scale-invariant feature transform, SIFT) extraction, index building, and image similarity scoring. We further propose a multi-index for SIFT descriptors, a secure bucket identifier computation protocol, and a secure image similarity computation protocol. In our framework, mobile users only need to encrypt and upload their images to the cloud. The later provides privacy-preserving image retrieval services without the mobile users extra interaction. Our framework greatly reduces computation overhead on the mobile user side. The security analysis and experimental evaluations demonstrate the efficiency and the scalability of our framework.