Research on Style Migration Techniques Based on Generative Adversarial Networks in Chinese Painting Creation

Abstract The continuous progress and development of science and technology have brought rich and diverse artistic experiences to the current society. The image style migration technology based on generative adversarial networks is more effective in meeting people’s specific artistic needs. However, the traditional algorithm model still cannot effectively meet the technical needs of style migration in Chinese painting creation, which puts new demands on the existing generative adversarial network model. To this end, this paper adopts CycleGAN generative adversarial network model, in-depth study of the loss function design ideas in the model, the introduction of ResNeSt network structure to optimize the algorithm of the generative adversarial network model, and the optimized model to carry out the effectiveness of experiments and user surveys. The optimized CycleGAN adversarial network model obtains the lowest score of 180.0012 in the index FID, the highest score of 0.9119 in the index SSIM, and the lowest score of 0.950 in the index Kernel MMD in the validity experiments. In the user survey, the optimized CycleGAN adversarial network model of this paper obtained the highest average score of 4.33 from university students, and the highest average score of 4.2 from experts. Compared with the other algorithmic models, the model of this paper is able to learn the style and features of Chinese painting creation, retain the original semantic information of the image, and realize the high-quality migration technology of Chinese painting creation style.