Scholarlink

<p>To address the challenge of complex texture transfer in ceramic decorative pattern design, this study proposes an automated style transfer algorithm combining an improved CycleGAN and Stable Diffusion model. Image features are decomposed into low-frequency structure and high-frequency detail via a frequency separation channel attention mechanism, enhancing texture correlation through weighted fusion. A lightweight VGG-16 with a mirrored streaming pyramid module enables multi-scale feature alignment and optimizes generator efficiency. Stable Diffusion dynamically adjusts generation parameters, improving controllability and artistic quality. Experiments show the algorithm outperforms traditional methods with a PSNR of 47.2&#xa0;dB and SSIM of 0.92, while preserving fine details. Pattern extraction accuracy exceeds 90%, and the F1 score reaches 85.16. This method effectively reduces texture blurring, balancing quality and efficiency, and provides an automated tool for digital ceramic design and intelligent cultural heritage preservation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated style transfer generation algorithm for ceramic decorative pattern design

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月28日

查看全文
  • Yichi Bi

摘要

To address the challenge of complex texture transfer in ceramic decorative pattern design, this study proposes an automated style transfer algorithm combining an improved CycleGAN and Stable Diffusion model. Image features are decomposed into low-frequency structure and high-frequency detail via a frequency separation channel attention mechanism, enhancing texture correlation through weighted fusion. A lightweight VGG-16 with a mirrored streaming pyramid module enables multi-scale feature alignment and optimizes generator efficiency. Stable Diffusion dynamically adjusts generation parameters, improving controllability and artistic quality. Experiments show the algorithm outperforms traditional methods with a PSNR of 47.2 dB and SSIM of 0.92, while preserving fine details. Pattern extraction accuracy exceeds 90%, and the F1 score reaches 85.16. This method effectively reduces texture blurring, balancing quality and efficiency, and provides an automated tool for digital ceramic design and intelligent cultural heritage preservation.