Privacy-Preserving System for Image Analytics in Cloud Environment Using Autoencoders
V Sivak K, Srinivasa Sai Abhijit Challapalli, Faheemullah Fathemullam, Vincent Tony Blair, Sri Lakshmi, Mahesh V · 2025
Privacy-preserving analytics of image data is important in the age of constant data breaches in cloud environments and cyber attacks. This paper introduces a new framework that makes use of an autoencoder-based convolutional neural network (CNN) to provide secure and efficient image analytics within cloud environments. The proposed system separates the model into two main parts, an encoder running on client devices as a public key and a pretrained CNN running on the cloud. The encoder extracts features with dimensionality reduction and privacy protection, ensuring raw image data is never accessible in the cloud server. This two-way benefit not only protects sensitive data but also improves storage efficiency. The cloud server carries out computationally intensive analysis on the encoded feature space with performance similar to conventional setup which requires raw data whereas this framework works without knowledge of the original image content. We benchmark our system on benchmark dataset CIFAR-10 and demonstrate that it maintains good classification accuracy while considerably enhancing privacy preservation and data compression. Our experiments prove the efficacy of this framework in balancing privacy, efficiency, and accuracy, which makes it a feasible solution for secure image-based analytics in the cloud environment.