Scholarlink

<p>Murals are a vital form of traditional Chinese art, rich in historical and cultural content. Line drawing, as a core technique, is widely used but still depends on manual tracing. This paper proposes an automatic method for generating mural line drawings by integrating edge enhancement, neural edge detection, and denoising. Enhance edges using image processing techniques, followed by a neural network (MLineNet) for line extraction. A cycle-consistent generative adversarial network (CycleGAN) refines the output by removing noise while preserving structural clarity. The model was evaluated using four metrics: structural similarity index (SSIM), texture complexity (TC), line connectivity index (LCI), and a comprehensive score (<i>Q</i>). On Dunhuang murals, it achieved scores of 89.54%, 93.77%, 88.14%, and 90.48%, respectively, and showed generalization to Baisha murals (<i>Q</i> = 89.29%). Results demonstrate the method’s reliability in producing complete, clean, and continuous mural line drawings.</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

Automatic generation of Chinese mural line drawings via enhanced edge detection and CycleGAN-based denoising

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月16日

查看全文
  • Haixia Feng,
  • Qingwu Hu,
  • Pengcheng Zhao,
  • Daoyuan Zheng,
  • Mingyao Ai,
  • Siliang Chen,
  • Xiyu Hu

摘要

Murals are a vital form of traditional Chinese art, rich in historical and cultural content. Line drawing, as a core technique, is widely used but still depends on manual tracing. This paper proposes an automatic method for generating mural line drawings by integrating edge enhancement, neural edge detection, and denoising. Enhance edges using image processing techniques, followed by a neural network (MLineNet) for line extraction. A cycle-consistent generative adversarial network (CycleGAN) refines the output by removing noise while preserving structural clarity. The model was evaluated using four metrics: structural similarity index (SSIM), texture complexity (TC), line connectivity index (LCI), and a comprehensive score (Q). On Dunhuang murals, it achieved scores of 89.54%, 93.77%, 88.14%, and 90.48%, respectively, and showed generalization to Baisha murals (Q = 89.29%). Results demonstrate the method’s reliability in producing complete, clean, and continuous mural line drawings.