Scholarlink

<p>Low-light image enhancement (LLIE) is a critical task in computer vision, aimed at restoring visibility and recovering fine details from images captured under poor illumination. However, existing LLIE methods often exhibit high sensitivity to brightness-color correlation, leading to undesired noise, color distortion, and visual artifacts. Moreover, many of them struggle to effectively preserve and enhance fine-grained structural details. To address these challenges, we propose a novel dual-stage LLIE framework, termed HIDENet, based on the latest Horizontal/Vertical-Intensity (HVI) color space. Specifically, HIDENet begins with a customized HVI color transformation that preliminarily decomposes the input image into Horizontal/Vertical (HV) and Intensity (I) maps. Then, the first stage (i.e., Multi-scale Attention Detail Extraction Stage) is designed to extract refined features from HV and I branches separately, while the second stage (i.e., Low-light Color Restoration Stage) is introduced to jointly integrate and fuse structural information from both branches, ensuring accurate color restoration and structural consistency. This dual-stage joint optimization pipeline not only mitigates the limitations caused by brightness–color entanglement but also enables more effective detail preservation and enhancement. Extensive experiments on multiple benchmark datasets demonstrate that HIDENet achieves competitive or superior performance compared to existing state-of-the-art methods in both qualitative and quantitative evaluations.</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

HIDENet: An HVI-based illumination and detail enhancement network for low-light images

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Yi Zhu,
  • Haoze Gao,
  • Bowen Li,
  • Wei Wang,
  • Mingming Zhang,
  • Fugui Xing,
  • Chao Xie

摘要

Low-light image enhancement (LLIE) is a critical task in computer vision, aimed at restoring visibility and recovering fine details from images captured under poor illumination. However, existing LLIE methods often exhibit high sensitivity to brightness-color correlation, leading to undesired noise, color distortion, and visual artifacts. Moreover, many of them struggle to effectively preserve and enhance fine-grained structural details. To address these challenges, we propose a novel dual-stage LLIE framework, termed HIDENet, based on the latest Horizontal/Vertical-Intensity (HVI) color space. Specifically, HIDENet begins with a customized HVI color transformation that preliminarily decomposes the input image into Horizontal/Vertical (HV) and Intensity (I) maps. Then, the first stage (i.e., Multi-scale Attention Detail Extraction Stage) is designed to extract refined features from HV and I branches separately, while the second stage (i.e., Low-light Color Restoration Stage) is introduced to jointly integrate and fuse structural information from both branches, ensuring accurate color restoration and structural consistency. This dual-stage joint optimization pipeline not only mitigates the limitations caused by brightness–color entanglement but also enables more effective detail preservation and enhancement. Extensive experiments on multiple benchmark datasets demonstrate that HIDENet achieves competitive or superior performance compared to existing state-of-the-art methods in both qualitative and quantitative evaluations.