Visually Image Encryption and Compression using a CNN-Based Autoencoder

This paper proposes a visual encryption method to ensure the confidentiality of digital images. The model used is based on an autoencoder using aConvolutional Neural Network (CNN) to ensure the protection of the user data on both the sender side (encryption process) and the receiver side(decryption process)in a symmetric mode. To train and test the model, we used the MNIST and CIFAR-10 datasets. Our focus lies in generating an encrypted dataset by combining the original dataset with a random mask. Then, a convolutional autoencoder in the masked dataset will be designed and trained to learn essential image features in a reduced-dimensional latent space and reconstruct the image from this space. The used mask can be considered as a secret key known in standard cryptographic algorithms which allows the receiver of the masked data to recover the plain data. The implementation of this proposed encryption model demonstrates efficacy in preserving data confidentiality and integrity while reducing the dimensionality (for example we pass from 3072 Bytes to 1024 Bytes for CIFAR-10 images). Experimental results show that the used CNN exhibits a proficient encryption and decryption process on the MNIST dataset, and a proficient encryption and acceptable decryption process on the CIFAR-10 dataset.