Wall art image stitching and generation based on mask-fine GAN model

In recent years, wall art has become an important part of urban culture. However, image generation and restoration technologies still have limitations, especially in terms of image structure coherence and the consistency of artistic style. Issues such as abrupt textures and discontinuous edges at image stitching points are prominent. To address these problems, this study proposes a wall art image stitching and generation method that integrates Mask Generative Adversarial Network and Fine Generative Adversarial Network. The final model design performs excellently on public image datasets. In terms of generated image quality, the proposed model achieves Frechet Inception Distance and initial scores of 8.2 and 8.6, respectively, outperforming comparative models. Moreover, the model’s no-reference image quality evaluation, measured using the BRISQUE index, reaches 11.8, indicating that the generated images have a certain degree of realism. The proposed improved model demonstrates stronger applicability and reliability in the task of wall art image stitching and generation. The findings of this study provide technical support for the digital restoration, artistic creation, and urban visual design of wall art. It can improve the automatic generation and restoration quality of wall art images, promote the development of wall art creation technology, and support its practical application in digital cultural industries and urban esthetic improvement.