Scholarlink

<p>Atmospheric dust particles significantly degrade image quality through scattering and absorption, hindering reliable computer vision applications such as intelligent transportation and remote sensing. Yet, current approaches face limitations due to the scarcity of paired dust image datasets and unstable performance under varying dust concentrations. To address these issues, this paper proposes a dust image clarification framework that combines hierarchical color cast correction with a Dual-Domain Feature Fusion Dehazing Network (DDFD-Net). A fused color shift metric guides a two-stage correction strategy—global mean correction for mild bias and convolution-based local mean correction with Gamma adjustment for severe cases. For deblurring, DDFD-Net enhances All-in-One Dehazing Network (AOD-Net) by integrating a detail-enhancing block and fusing spatial and frequency-domain features via Fast Fourier Transformation (FFT) and a Squeeze-and-Excitation (SE) mechanism. Training is conducted on a haze-like dataset synthesized from NYUv2 images and further validated on a large set of real-world dust images. Experimental results confirm that the proposed method achieves clearer, more natural restorations than state-of-the-art algorithms, demonstrating strong robustness and practical potential in dust-affected vision systems.</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

Enhanced dust image clarification via hierarchical color cast correction and dual–domain feature fusion

  • 期刊论文

  • 发表日期：

    2026年7月18日

查看全文
  • Hongxia Niu,
  • Zuoyi Xie,
  • Tao Hou

摘要

Atmospheric dust particles significantly degrade image quality through scattering and absorption, hindering reliable computer vision applications such as intelligent transportation and remote sensing. Yet, current approaches face limitations due to the scarcity of paired dust image datasets and unstable performance under varying dust concentrations. To address these issues, this paper proposes a dust image clarification framework that combines hierarchical color cast correction with a Dual-Domain Feature Fusion Dehazing Network (DDFD-Net). A fused color shift metric guides a two-stage correction strategy—global mean correction for mild bias and convolution-based local mean correction with Gamma adjustment for severe cases. For deblurring, DDFD-Net enhances All-in-One Dehazing Network (AOD-Net) by integrating a detail-enhancing block and fusing spatial and frequency-domain features via Fast Fourier Transformation (FFT) and a Squeeze-and-Excitation (SE) mechanism. Training is conducted on a haze-like dataset synthesized from NYUv2 images and further validated on a large set of real-world dust images. Experimental results confirm that the proposed method achieves clearer, more natural restorations than state-of-the-art algorithms, demonstrating strong robustness and practical potential in dust-affected vision systems.