Scholarlink

<p>In nighttime photography, lens flare caused by internal reflection and scattering of light within the camera lens often degrades image quality. Existing methods struggle to effectively remove both reflective and scattering flares while preserving original image details. This paper proposes a physically-inspired frequency decomposition and fusion network for robust nighttime flare removal. The method converts the image to the HSI color space and separates the intensity component. A Fourier Decomposition Module then decomposes this component into a low-frequency part containing global illumination and a high-frequency part containing detail information. A Uformer-based Flare Removal Module suppresses low-frequency flare, and a Multi-Scale Fusion Module reconstructs the flare-free image. Experiments on the Flare7K dataset demonstrate the superior performance of our method, achieving a PSNR of 27.12 dB and an SSIM of 0.903 on real-world scene images, effectively accomplishing flare removal and detail preservation.</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

Physically-inspired frequency decomposition and fusion network for robust nighttime flare removal

  • 期刊论文

  • 发表日期：

    2025年10月31日

查看全文
  • Zerun Zhang,
  • Yunqing Liu,
  • Fei Yan,
  • Dongpo Xu

摘要

In nighttime photography, lens flare caused by internal reflection and scattering of light within the camera lens often degrades image quality. Existing methods struggle to effectively remove both reflective and scattering flares while preserving original image details. This paper proposes a physically-inspired frequency decomposition and fusion network for robust nighttime flare removal. The method converts the image to the HSI color space and separates the intensity component. A Fourier Decomposition Module then decomposes this component into a low-frequency part containing global illumination and a high-frequency part containing detail information. A Uformer-based Flare Removal Module suppresses low-frequency flare, and a Multi-Scale Fusion Module reconstructs the flare-free image. Experiments on the Flare7K dataset demonstrate the superior performance of our method, achieving a PSNR of 27.12 dB and an SSIM of 0.903 on real-world scene images, effectively accomplishing flare removal and detail preservation.