Application of Neural Network Models in Image Encryption and Differential Analysis

As the demand for encryption and information security increases, the concept of neural networks in deep learning is gradually used in the field of image encryption, and the influence of neural networks on the field of image encryption is gradually deepening. The current mainstream neural network image encryption schemes are categorized into pixel disruption and chaotic systems. This paper provides a basic introduction to the four network models of chaotic neural networks, convolutional neural networks, cellular neural networks, and generative adversarial networks for image encryption, an analysis of the algorithmic framework, and an analysis of the encrypted image. It is found that all four neural network models are affected by the parameters of the model as well as the size, if the neural network encrypted image is highly resistant to noise, at the same time the distortion of the image will be larger and the visual entropy of the image will be increased accordingly. If a model is sensitive to the initial value, then the model has a relatively large key space, low correlation of neighboring pixels, good encryption, and is more difficult to crack.