Application of CycleGAN-based image style transfer algorithm in visual communication design

This paper addresses the challenges in traditional image style transfer research, including high design costs, reliance on paired data, limited transfer effects, and a lack of optimization for practical applications. To overcome these limitations, we introduce the CycleGAN algorithm for efficient image style transfer without paired data, aiming to enhance the efficiency, personalization, and diversity required in visual communication design. Using a pre-trained CycleGAN model, we conduct experiments with datasets obtained from DesignNet and Google Open Images, following a comprehensive preprocessing workflow. The effectiveness of the CycleGAN-based style transfer is quantified using the Structural Similarity Index (SSIM). Experimental results demonstrate that the CycleGAN model performs excellently in style transfer, achieving SSIM values of 0.85, 0.84, and 0.87 for classical painting, modern art, and retro poster styles, respectively. Further validation through user surveys and expert evaluations confirms the high quality of the generated images in terms of aesthetics, clarity, and recognizability, with a creativity score of 4.5 and a satisfaction rate of 90%. This study not only enriches the technical toolkit for visual communication design but also provides valuable insights into the application potential of CycleGAN, particularly in style transfer tasks that do not require paired samples, showcasing the algorithm’s unique advantages and practical applicability.