Scholarlink

<p>In recent years, significant progress has been made in real-world low-light image enhancement. However, most existing methods heavily rely on well-lit training data, limiting their generalization ability in unseen scenarios. Therefore, reducing the dependence on supervised signals has become a major challenge in low-light image enhancement. To address this issue, this paper proposes a novel zero-reference low-light image enhancement method that leverages the intrinsic illumination priors of images to achieve adaptive enhancement. Specifically, we thoroughly investigate the impact of wavelet transform on image illumination variations and discover that, after multiple transformations, the low-frequency components exhibit distinct illumination enhancement characteristics. Based on this observation, we incorporate the Retinex theory to construct a preliminary illumination enhancement network, which extracts and enhances the illumination information of the image. Furthermore, we design a Transformer-based light transfer network. It extracts content structures from the initially enhanced image, while capturing illumination backgrounds from the low-frequency domain obtained through multi-level wavelet decompositions. Extensive experimental results demonstrate that the proposed method exhibits excellent interpretability, robustness, and efficiency across various low-light scenarios, significantly improving the visual quality of low-light images. The code will be available at <a href="https://github.com/sxk2137/Wavelet">https://github.com/sxk2137/Wavelet</a>.</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

Wavelet transform-guided transformer light transfer network for zero-shot low-light image enhancement

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Minglong Xue,
  • Xukun Shang,
  • Jinhong He,
  • Senming Zhong

摘要

In recent years, significant progress has been made in real-world low-light image enhancement. However, most existing methods heavily rely on well-lit training data, limiting their generalization ability in unseen scenarios. Therefore, reducing the dependence on supervised signals has become a major challenge in low-light image enhancement. To address this issue, this paper proposes a novel zero-reference low-light image enhancement method that leverages the intrinsic illumination priors of images to achieve adaptive enhancement. Specifically, we thoroughly investigate the impact of wavelet transform on image illumination variations and discover that, after multiple transformations, the low-frequency components exhibit distinct illumination enhancement characteristics. Based on this observation, we incorporate the Retinex theory to construct a preliminary illumination enhancement network, which extracts and enhances the illumination information of the image. Furthermore, we design a Transformer-based light transfer network. It extracts content structures from the initially enhanced image, while capturing illumination backgrounds from the low-frequency domain obtained through multi-level wavelet decompositions. Extensive experimental results demonstrate that the proposed method exhibits excellent interpretability, robustness, and efficiency across various low-light scenarios, significantly improving the visual quality of low-light images. The code will be available at https://github.com/sxk2137/Wavelet.