Scholarlink

<p>Image and video dehazing are essential tasks in Image Reconstruction, crucial for restoring the visibility and clarity of scenes obscured by haze, fog, or smoke. Unlike conventional methods, which often struggle with preserving fine details and color fidelity, the proposed model Hybrid IA-RDUNet excels in enhancing image quality by leveraging advanced feature extraction and reconstruction techniques. The proposed model combines the encoder-decoder structure of UNet with the advanced feature extraction capabilities of Residual Dense Networks (RDN), augmented by Intensity Attention Blocks to emphasize critical features and Sub-Pixel Upsampling for precise image reconstruction. Comprehensive evaluations on benchmark datasets, including REVIDE, NH-HAZE, SOTS, D-Hazy, and I-Haze, reveal that our approach significantly outperforms existing methods in terms of PSNR, SSIM, and color accuracy. The Hybrid IA-RDUNet model offers a robust, efficient, and perceptually superior solution for enhancing visual content under hazy conditions.</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

Hybrid IA-RDUNet: a novel approach to image and video dehazing

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Parth Chande,
  • Gajanan Galshetwar,
  • Vijay Galshetwar,
  • Niteshkumar Agrawal,
  • Neha Mishra

摘要

Image and video dehazing are essential tasks in Image Reconstruction, crucial for restoring the visibility and clarity of scenes obscured by haze, fog, or smoke. Unlike conventional methods, which often struggle with preserving fine details and color fidelity, the proposed model Hybrid IA-RDUNet excels in enhancing image quality by leveraging advanced feature extraction and reconstruction techniques. The proposed model combines the encoder-decoder structure of UNet with the advanced feature extraction capabilities of Residual Dense Networks (RDN), augmented by Intensity Attention Blocks to emphasize critical features and Sub-Pixel Upsampling for precise image reconstruction. Comprehensive evaluations on benchmark datasets, including REVIDE, NH-HAZE, SOTS, D-Hazy, and I-Haze, reveal that our approach significantly outperforms existing methods in terms of PSNR, SSIM, and color accuracy. The Hybrid IA-RDUNet model offers a robust, efficient, and perceptually superior solution for enhancing visual content under hazy conditions.