Scholarlink

<p>Image restoration aims to recover high-quality images from degraded inputs. Traditional methods often struggle with complex, real-world scenarios involving multiple simultaneous degradations. We propose MCDRNet, a multi-granularity approach for all-in-one image restoration via contrast-guided degradation reconstruction, to address this challenge. MCDRNet utilizes a Contrastive Degradation Representation Extractor (CDRE) to capture shallow degradation features and a Multi-Granularity Collaborative Attention Module (MCAM) to integrate multi-scale semantic information for deep feature extraction. An Adaptive Restoration Network (ARN) then reconstructs high-quality images by decoupling degradation and content features. Extensive experiments on three datasets demonstrate MCDRNet’s superior performance over 22 baselines in denoising, deraining, and dehazing tasks. The code and datasets are available at <a href="https://github.com/HyaLii/MCDRNet.git">https://github.com/HyaLii/MCDRNet.git</a>.</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

MCDRNet: a multi-granularity approach for all-in-one image restoration via contrast-guided degradation reconstruction

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • Yali Han,
  • Jiqiang Tang

摘要

Image restoration aims to recover high-quality images from degraded inputs. Traditional methods often struggle with complex, real-world scenarios involving multiple simultaneous degradations. We propose MCDRNet, a multi-granularity approach for all-in-one image restoration via contrast-guided degradation reconstruction, to address this challenge. MCDRNet utilizes a Contrastive Degradation Representation Extractor (CDRE) to capture shallow degradation features and a Multi-Granularity Collaborative Attention Module (MCAM) to integrate multi-scale semantic information for deep feature extraction. An Adaptive Restoration Network (ARN) then reconstructs high-quality images by decoupling degradation and content features. Extensive experiments on three datasets demonstrate MCDRNet’s superior performance over 22 baselines in denoising, deraining, and dehazing tasks. The code and datasets are available at https://github.com/HyaLii/MCDRNet.git.