Research on Digital Reproduction and Visual Communication of Intangible Cultural Heritage Crafts Integrating Generative Adversarial Networks

Intangible cultural heritage (ICH) crafts carry rich historical and cultural value, but traditional digitization methods struggle to accurately reproduce complex textures and craft details, and their visual immersion is limited. This paper proposes a model for the digital reproduction and visual communication of ICH crafts based on a generative adversarial network (GAN). Through multimodal feature fusion and a conditional generation mechanism, this model achieves high-precision, style-consistent craft image generation. Experimental results show that texture restoration achieves PSNRs of 28.5 dB, 29.8 dB, and 27.6 dB for embroidery, ceramics, and woodcarving, respectively, and SSIMs of 0.89, 0.91, and 0.87, respectively. Style transfer achieves a minimum LPIPS value of 0.098 and a maximum color matching rate of 93.2%, demonstrating excellent visual fidelity and color reproduction for the generated images. The model also demonstrates good generalization to new craft types, with a stable generation time of 0.4 to 0.45 seconds. User interaction experience surveys indicate an average interest match score of 4.37 and an average immersion score of 4.21. The research results demonstrate that GAN-based digitization methods can efficiently reproduce the details of intangible cultural heritage crafts, providing reliable technical support for cultural communication, educational display, and personalized digital experiences.