Research on Image Encryption Algorithm Based on Generative Neural Network

The security of digital image transmission is related to the national economy and people’s livelihood, and the traditional text encryption algorithm performs poorly on images. Based on this, this paper designs the image encryption model, optimizes the generator and discriminator network structure based on the residual network model, and improves the generative adversarial network. Logistic chaotic system is utilized to construct a new GAN key generation model, and it is applied to the field of image encryption. On this basis, the Knuth-Durstenfeld algorithm is improved to complete the encryption of images from three processes: pixel value replacement, image disruption, and image diffusion, and applied research is carried out. The results show that the histograms of Lena and Peppers plaintext images show irregular statistical characteristics, and the gray value distribution pattern of the histograms of ciphertext images has no big ups and downs and tends to be stable and smooth. The correlation coefficients of adjacent pixels of ciphertext image are close to the ideal value of 0, and the image information is more difficult to be analyzed and deciphered. Under the noise attack, the PSNR values of the ciphertext image do not differ much, which can prove that the ability of anti-noise attack is good. The encryption algorithm based on generative adversarial network shows excellent performance on image encryption and protects the information security of digital images.