Efficient & Secure similarity search for encrypted images over cloud

Abstract

With the expansion of edge devices, the Content-Based Image Retrieval (CBIR) technique has attracted attention across sectors like distributed computing, peer-to-peer communication networks. Nevertheless, existing CBIR strategies aimed at image security and retrieval assistance exhibit certain inherent limitations including subpar search precision, restricted search capabilities, and susceptibility to key exposure. To address these concerns, our framework proposes a similar search methodology suited for secure distributed computing employing encrypted images. Data owners initially encrypt their images through keys generated by a shared key management tool. Subsequently, features are extracted via a pre-trained CNN model, ensuring security, while secure indices are constructed employing the K-Nearest Neighbors [KNN] algorithm to bolster search accuracy. The encrypted images are then transmitted to the cloud. This model guarantees heightened search precision and efficiency, simultaneously thwarting key compromise and enhancing overall image security.