Research on Traditional Pattern Generation Based on Adaptive Fusion Attention Mechanism

This is the authors' abstract. We don't add key points for this paper.

Traditional patterns, characterized by their intricate designs and vibrant colors, hold significant application value for generating cultural and artistic images that incorporate national characteristics while preserving the original pattern structure. To achieve high-quality traditional pattern image generation, this study introduces a novel method based on adaptive fusion attention. The AdaAttN module is integrated into the generator to enhance visual focus and improve image quality. A multi-scale feature fusion discriminator is employed to accurately distinguish between the global and local features of the patterns, thereby perfecting the pattern stylization. The training process is guided by a stylistic feature fusion loss function, which ensures that the generated images reflect more ethnic colors and styles. Experimental comparative analysis conducted on a real dataset demonstrates that the model enhances image generation quality, confirming its feasibility and artistic merit.