Traditional Chinese patterns encounter challenges in contemporary textile design in terms of low production efficiency, an insufficient amount of precision, and a lack of innovation. In response, this paper designs a digital patterned textiles generation system. Specifically, this system extracts geometric structures and colors using a convolutional neural net (CNN) and a residual net. The system implements a conditional generative adversarial network (cGAN) that generates features based on a combined feature vector and conditional information during generation to ensure consistency of style. Finally, an adaptive style transfer algorithm is used to fine-tune lines, colors, and textures by optimizing a combination of content loss, style loss, and edge preservation loss. The patterns are finally enhanced through high-frequency reconstruction using a super-resolution convolutional neural net (SRCNN) to aid clarity and high-resolution patterns. The experimental results show the generated pattern achieves a structural similarity (SSIM) index of 0.99, a peak signal-to-noise ratio (PSNR) of 37 dB, a sharpness index of 0.95, and accuracy of geometric elements representation in cultural symbols with up to 98% accuracy. This work successfully enables the innovative integration of the digital inheritance of traditional patterns and contemporary textile design.

digital design, traditional patterns, generative adversarial network, image processing, cultural symbols