Scholarlink

<p>Image-capturing limitations and environmental factors often degrade image quality, particularly in low-light conditions, which affects both the visual appeal and the performance of computer vision algorithms. Enhancing low-light images can help retrieve useful information and improve visual quality. The proposed method follows three main steps: (1) compute entropy for each gray level and plot the entropy distribution, (2) update the distribution using a uniform reference and equalize it, and (3) apply homomorphic filtering to reduce noise. Finally, adaptive gamma correction with a weighted distribution (AGCWD) enhances contrast and overall quality for grayscale and color images. The method’s effectiveness is assessed through qualitative and quantitative analysis using the LOL dataset. Its performance is benchmarked against leading techniques such as WAHE, ROHIM, ESIHE, QDHE, BIMEF, DnCNN, MSRCR, DWTSVD, OHCICD, and FUZZY DCT. Evaluation metrics include NIQMC, PCQI, RCM, CEIQ, and EBCM, confirming the method’s viability for real-world image processing applications.</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

Advancing low-light image enhancement: an entropy-driven approach with reduced defects

  • 期刊论文

  • 发表日期：

    2025年7月23日

查看全文
  • Nitish Kumar,
  • Ravi Kumar,
  • Syed Shahnawazuddin

摘要

Image-capturing limitations and environmental factors often degrade image quality, particularly in low-light conditions, which affects both the visual appeal and the performance of computer vision algorithms. Enhancing low-light images can help retrieve useful information and improve visual quality. The proposed method follows three main steps: (1) compute entropy for each gray level and plot the entropy distribution, (2) update the distribution using a uniform reference and equalize it, and (3) apply homomorphic filtering to reduce noise. Finally, adaptive gamma correction with a weighted distribution (AGCWD) enhances contrast and overall quality for grayscale and color images. The method’s effectiveness is assessed through qualitative and quantitative analysis using the LOL dataset. Its performance is benchmarked against leading techniques such as WAHE, ROHIM, ESIHE, QDHE, BIMEF, DnCNN, MSRCR, DWTSVD, OHCICD, and FUZZY DCT. Evaluation metrics include NIQMC, PCQI, RCM, CEIQ, and EBCM, confirming the method’s viability for real-world image processing applications.