Lightweight and Controllable Privacy-Preserving Image Retrieval in Multi-User Settings

In cloud environments, privacy-preserving contentbased image retrieval (PPCBIR) enables users to retrieve images while protecting image privacy. Existing PPCBIR systems often use a single image key, which causes low efficiency and makes it difficult to achieve fine-grained access control over images. This paper proposes a lightweight and controllable privacy-preserving image retrieval in multi-user settings (named LCPIRM) to improve time efficiency and access control performance. A one-time image encryption method based on reversible embedding is proposed to balance the contradiction between complexity and security without increasing the difficulty of key management. A robust hash generation method is designed by combining piecewise mean quantization and encryption image features, which can effectively improve retrieval efficiency because the robust hashes embedded in the encrypted images can be extracted and establish inverted indexing in the cloud. When dealing with authorized encrypted images, the cloud server uses proxy re-encryption to convert the image keys embedded within themselves from the owner's public key protection to the authorized user's public key protection, achieving fine-grained access control over images in a multi-user setting. Theoretical analysis and experimental results show that LCPIRM has better performance in terms of retrieval accuracy, consumption, and search efficiency while meeting security requirements. In the real datasets Caltech256 and Caltech101, the search efficiency has increased by 74% and 58% respectively compared to the existing schemes.