SEEIR: Secure and Efficient Encrypted Image Retrieval Based on Additive Secret Sharing
Yangyu Shen, Haijiang Wang, Jian Wan, Lei Zhang, Huang J. Jie, Zegang Pan · IEEE Transactions on Network Science and Engineering · 2025
Content-based image retrieval (CBIR) leverages convolutional neural networks (CNN) (e.g., VGG-16) to achieve high accuracy by extracting image feature vectors. While existing schemes employ additive secret sharing (ASS) with a twin-cloud model to delegate tasks like secure feature extraction and encrypted retrieval to cloud servers, they suffer from critical limitations: (1) insecure index structures vulnerable to unauthorized queries and (2) inefficient twin-server communication protocols. To address these issues, we propose SEEIR, a secure and efficient encrypted image retrieval scheme based on ASS. First, SEEIR enhances retrieval security through secure kNN-ASS, a novel method that encrypts index shares across twin clouds to enforce access control. Only users with keys authorized by the data owner can generate valid query vectors, blocking adversarial attempts to compromise sensitive metadata. Second, SEEIR eliminates the twin-server communication overhead by securely merging encrypted index shares into a single cloud server, improving retrieval efficiency. Finally, both theoretical analysis and empirical experiments confirm the security and efficiency of the proposed scheme.