The recovery scheme of improved double random phase coding encryption-hidden images based on deep learning and chaos

Abstract

In order to improve the security of ciphertext image in transmission, a scheme of using neural network is proposed to restore the encryption-hiding images by using chaotic iris phase mask and double random phase encoding encrypted. To improve encryption performance, we replace the double random phase with chaotic iris phase masks. Firstly, the plaintext image is encrypted by improved double chaotic iris phase masks encoding, and the ciphertext image is generated. Then the ciphertext image is hidden into the carrier image through the Least Significant Bit (LSB) algorithm. It can avoid the detection to a certain extent. And a large number of hidden-plaintext image pairs are produced as a dataset. Secondly, the neural network is built in the process of continuous training and testing. The neural network selected in this paper is residual network (ResNet). The neural network to learn the logical relationship between input and output more efficiently, and successfully recovered the corresponding plaintext image. Finally, the trained neural network can fit the mapping relationship between the hidden image and the plaintext image. When decrypting, it is not necessary to extract the ciphertext image from the hidden image first and then decrypt it. This scheme can directly realize the decryption of hidden images. The paper elaborates the encryption, hiding and decryption process of the scheme in detail. Simulation experiments show that the scheme is feasible and has good robustness.