Secure Content Based Image Retrieval System Using Deep Learning

Abstract

This paper investigates Content-Based Image Retrieval (CBIR) using an ensemble of three cutting-edge deep learning architectures: Xception, MobileNet, and Inception. This ensemble approach demonstrated exceptional retrieval accuracy, with Xception and Inception models achieving an accuracy of 92.375%, precision and recall of 93% and 92% respectively, and an F1-score of 92%. The MobileNet model also showed strong performance, with an accuracy of 87.125%, precision and recall of 88% and 87%, and an F1-score of 87%.Beyond mere retrieval accuracy, the study places a significant emphasis on the security of the image database. A dual-layer encryption method was employed, integrating visual cryptography with the Advanced Encryption Standard (AES) to ensure robust protection of sensitive data. This approach guarantees efficient image retrieval based on content while securing the data against potential breaches.The results underscore the efficiency of the ensemble model in balancing high retrieval accuracy with stringent security measures. This balance is particularly relevant for applications in digital libraries, historical research, fingerprint identification, and crime prevention. The paper’s findings advocate for the critical need to integrate strong security protocols in future CBIR systems, ensuring optimal performance without compromising data security.