Scholarlink

<p>The integration of spatial and frequency domains offers innovative solutions for low-light image enhancement (LLIE). However, most existing methods rely on independent branches for feature extraction, which often lacks deep interaction and effective fusion between the two domains. To address this, we propose SynerFormer, a novel spatial-frequency synergistic framework designed for robust enhancement. To improve the network’s perception of complex lighting conditions, we first introduce a CNN-based Illumination Component Estimator (LCP) to extract illumination priors and generate guidance signals. Furthermore, to overcome the difficulty of effectively coordinating dual-domain features, we design the core Spatial-Frequency Synergistic Attention (SFSA) module. This module decouples features into spatial and frequency branches, facilitates deep interaction via a symmetric attention mechanism, and employs LCP guidance signals for dynamic feature modulation prior to fusion. By integrating the SFSA into a U-Net architecture, we construct the Synergistic Restoration (SR) module, which enhances local textures while ensuring global luminance and chromatic consistency. Extensive experiments demonstrate that the proposed method achieves superior performance in terms of both subjective visual quality and objective evaluation metrics.</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

A Spatial–Frequency Synergistic Attention Transformer Network for Low-Light Image Enhancement

  • 期刊论文

  • 发表日期：

    2026年5月24日

查看全文
  • Yumei Zhan,
  • Haiyan Quan,
  • Lisha Zou,
  • Zhewen Ni,
  • Zongxi Wang,
  • Yuqin He

摘要

The integration of spatial and frequency domains offers innovative solutions for low-light image enhancement (LLIE). However, most existing methods rely on independent branches for feature extraction, which often lacks deep interaction and effective fusion between the two domains. To address this, we propose SynerFormer, a novel spatial-frequency synergistic framework designed for robust enhancement. To improve the network’s perception of complex lighting conditions, we first introduce a CNN-based Illumination Component Estimator (LCP) to extract illumination priors and generate guidance signals. Furthermore, to overcome the difficulty of effectively coordinating dual-domain features, we design the core Spatial-Frequency Synergistic Attention (SFSA) module. This module decouples features into spatial and frequency branches, facilitates deep interaction via a symmetric attention mechanism, and employs LCP guidance signals for dynamic feature modulation prior to fusion. By integrating the SFSA into a U-Net architecture, we construct the Synergistic Restoration (SR) module, which enhances local textures while ensuring global luminance and chromatic consistency. Extensive experiments demonstrate that the proposed method achieves superior performance in terms of both subjective visual quality and objective evaluation metrics.