Visually Image Encryption based on Efficient Deep Learning Autoencoder


Mohamed Abdelmalek, Anis Harhoura, Issam Elaloui, Mahdi Madani and El-Bay Bourennane, University of Burgundy, France


This paper proposes an Artificial Intelligence (AI) model based on Convolutional Neural Network (CNN) for visual image protection during encryption and decryption processes. We used the CIFAR-10 dataset containing 60.000 color images of size 32x32 across ten classes to train and test the proposed network. Our focus lies in designing a convolutional autoencoder for image compression and reconstruction, utilizing an encoder-decoder architecture. During training, the autoencoder learns to encode essential image features into a reduced-dimensional latent space and reconstructs the image from this space. The implementation of the proposed encryption model demonstrates efficacy in preserving data integrity while reducing dimensionality. Experimental results show that the used CNN exhibits a proficient encryption process and acceptable decryption process.


Visually image protection, Deep Learning, Encryption, Encoder, Decoder, Data security, Image Compression.

Full Text  Volume 14, Number 12