Scholarlink

<p>Jiandu, ancient China’s primary writing medium before paper, frequently exhibits severe inscription degradation due to prolonged burial, making the texts difficult to decipher. To enhance textual visibility, this study proposes an adaptive infrared-visible image fusion method driven by Salient Spatial Attention (SSA), integrating the complementary strengths of both modalities to produce images with clear ink, texture, and color. First, an SSA module selectively enhances infrared ink features via adaptive feature selection, improving the visibility of faint inscriptions. Second, a multi-scale information measurement strategy ensures balanced representation of material texture and ink details. Finally, a tailored unsupervised loss function eliminates reliance on ground-truth images while preserving authentic visual characteristics. Experimental results demonstrate that the proposed approach outperforms existing methods, particularly in preserving text details, material texture, and color fidelity. The resulting enhanced images offer valuable support for cultural-heritage preservation and Jiandu research.</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

SSA-based adaptive infrared-visible image fusion for ink enhancement in ancient bamboo slips

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月13日

查看全文
  • Qiang Zhang,
  • Jiazhen Qin,
  • Ying Qi,
  • Teng Wan,
  • Shanxiong Chen,
  • Lixin Yang,
  • Chenyang Wang,
  • Xin Zhang,
  • Jintan Han,
  • Fengchen Qi

摘要

Jiandu, ancient China’s primary writing medium before paper, frequently exhibits severe inscription degradation due to prolonged burial, making the texts difficult to decipher. To enhance textual visibility, this study proposes an adaptive infrared-visible image fusion method driven by Salient Spatial Attention (SSA), integrating the complementary strengths of both modalities to produce images with clear ink, texture, and color. First, an SSA module selectively enhances infrared ink features via adaptive feature selection, improving the visibility of faint inscriptions. Second, a multi-scale information measurement strategy ensures balanced representation of material texture and ink details. Finally, a tailored unsupervised loss function eliminates reliance on ground-truth images while preserving authentic visual characteristics. Experimental results demonstrate that the proposed approach outperforms existing methods, particularly in preserving text details, material texture, and color fidelity. The resulting enhanced images offer valuable support for cultural-heritage preservation and Jiandu research.