Scholarlink

<p>In digital printing, the conversion from device-independent CIELAB color space to device-dependent CMYK space is crucial for color management. Traditional methods, such as Gray Component Replacement (GCR) and Under Color Removal (UCR), often rely on fixed rules and struggle with modeling complex CMYK channel interactions. Here, we introduce an unsupervised deep learning framework, MCU-Net, which performs Lab-to-CMYK mapping without explicit CMYK supervision. By incorporating differentiable constraint networks that reflect physical properties, such as perceptual color consistency and feasible K-range prediction, MCU-Net guides the main network to generate low-ink, color-accurate outputs. Experimental results demonstrate an average ink saving of 37.13% and a mean <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\Delta E_{2000}\)</EquationSource> </InlineEquation> of 0.38 in real printed samples, outperforming traditional GCR and commercial software. This framework offers a practical approach for ICC profile generation, balancing ink efficiency and color fidelity in industrial color separation tasks.</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

Efficient ink-saving lab-to-CMYK color mapping through unsupervised deep learning with physical constraints

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • Yinwei Zhang,
  • Hongwu Zhan,
  • Libin Zhang,
  • Fang Xu,
  • Yifei Zou

摘要

In digital printing, the conversion from device-independent CIELAB color space to device-dependent CMYK space is crucial for color management. Traditional methods, such as Gray Component Replacement (GCR) and Under Color Removal (UCR), often rely on fixed rules and struggle with modeling complex CMYK channel interactions. Here, we introduce an unsupervised deep learning framework, MCU-Net, which performs Lab-to-CMYK mapping without explicit CMYK supervision. By incorporating differentiable constraint networks that reflect physical properties, such as perceptual color consistency and feasible K-range prediction, MCU-Net guides the main network to generate low-ink, color-accurate outputs. Experimental results demonstrate an average ink saving of 37.13% and a mean \(\Delta E_{2000}\) of 0.38 in real printed samples, outperforming traditional GCR and commercial software. This framework offers a practical approach for ICC profile generation, balancing ink efficiency and color fidelity in industrial color separation tasks.