Innovative Application of Chinese Landscape Culture Painting Elements in AI Art Generation

In recent years, the swift advancement of artificial intelligence (AI) has significantly propelled the capabilities of data generation, a pivotal technology underpinning AI development. This has led to remarkable innovations across various domains, including images, text, and speech. In this study, we introduce a sophisticated image translation network tailored for AI-driven art generation of landscape painting elements. This network is enhanced by integrating a self-attention mechanism with orthogonal Jacobi regularization, aiming to elevate the quality of the generated artwork. Furthermore, we developed a landscape style transfer model using the image above translation network to encapsulate elements of Chinese landscape cultural paintings. This model emulates the artistic process employed by human artists, effectively capturing the essence of the content and infusing the intended style. The efficacy of the generated images is quantitatively assessed through Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) values. Comparative analysis of the style transfer outcomes across three models reveals that the Generative Adversarial Network (GAN) model exhibits the lowest PSNR and SSIM values. Although the CycleGAN model achieves commendable style transfer, its visual impact remains less pronounced in comparison to the Wasserstein GAN (WGAN). Subjective evaluations of image quality and style fidelity between the generated images and the input stylized images indicate that the WGAN outperforms other models. It achieves an average content evaluation score of 4.11 and a style score of 4.2 across nine style transformations, surpassing the average scores obtained by competing models.