A time-varying image encryption algorithm driven by neural network

Hangming Zhang

doi:10.1016/j.optlastec.2025.112751

Return

A time-varying image encryption algorithm driven by neural network

2025-06-12

PRE

Hangming Zhang

Hanping Hu *

Weiping Ding

DOI：10.1016/j.optlastec.2025.112751

Original

Original request for help

Save

Abstract

En 中文

Nowadays, the application of neural networks, genetic operations, and chaos to image encryption algorithms has become a focused area of frontier research. The neuronal mechanism and deep neural network have good nonlinear characteristics and meet the anti-aggression demands for cryptosystem security. What cannot be ignored is that due to multiplications between matrices in the computation of neuronal mechanisms and deep neural networks, the activation function cannot precisely recover the image, and the direct application of image encryption by neurons and deep neural networks cannot obtain secure results. Moreover, the performance of the adopted chaotic systems also determines the safety level of the cryptosystem. The fixed coupling structures in current research make the dynamic behavior easy to predict, bringing vulnerability to cryptosystems. Considering the aforesaid flaws, an image encryption algorithm using the neural-network pattern, genetic algorithm, and time-varying coupling structure of the chaotic system is proposed, helping improve the security and efficiency of our cryptosystem. Firstly, a time-varying coupled chaotic system is proposed, and highly stochastic sequences are used as keystreams. Then, the keystreams are processed using the neural-network mode and genetic operation to complete scrambling and diffusing. Ultimately, the simulations and security analyses indicate that our method reaches a highly safe level, which shows high efficiency and strong security, and meets the demands of various security applications.

Keywords:

Neural network

Nerve cell

Genetic operation

Convolution lattice

Chaos

Image encryption