Scholarlink

<p>Haze can seriously reduce image quality and affect tasks such as object detection and semantic segmentation. Existing dehazing methods often perform well on synthetic images but struggle with real-world hazy images, particularly those with non-uniform haze distribution. To address this challenge, we propose a Residual Channel Prior-Guided Multi-Scale Progressive Dehazing Network (MPDNet). MPDNet leverages the rich structural information contained in the Residual Channel Prior (RCP) of hazy images and introduces a Prior-Guided Block (PGB) to extract RCP maps at different dehazing stages. To better apply prior knowledge at different stages, we design a progressive dehazing network. The Attention-Guided Feature Memory module (AFM) aims to explore the correlation between current input and historical information and achieve cross-stage feature transfer. In addition, Multi-Scale Dehazing Unit (MDU) combining a Multi-Scale Feature Extraction Module (MSFEM) and a hybrid attention mechanism is used to restore the haze-free image. Extensive experiments demonstrate that MPDNet achieves state-of-the-art performance on both synthetic and real datasets, particularly excelling in handling non-uniform hazy images.</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

Residual channel prior-guided multi-scale progressive dehazing network with hybrid attention

  • 期刊论文

  • 发表日期：

    2025年3月16日

查看全文
  • Yiming Xing,
  • Jindong Zhang

摘要

Haze can seriously reduce image quality and affect tasks such as object detection and semantic segmentation. Existing dehazing methods often perform well on synthetic images but struggle with real-world hazy images, particularly those with non-uniform haze distribution. To address this challenge, we propose a Residual Channel Prior-Guided Multi-Scale Progressive Dehazing Network (MPDNet). MPDNet leverages the rich structural information contained in the Residual Channel Prior (RCP) of hazy images and introduces a Prior-Guided Block (PGB) to extract RCP maps at different dehazing stages. To better apply prior knowledge at different stages, we design a progressive dehazing network. The Attention-Guided Feature Memory module (AFM) aims to explore the correlation between current input and historical information and achieve cross-stage feature transfer. In addition, Multi-Scale Dehazing Unit (MDU) combining a Multi-Scale Feature Extraction Module (MSFEM) and a hybrid attention mechanism is used to restore the haze-free image. Extensive experiments demonstrate that MPDNet achieves state-of-the-art performance on both synthetic and real datasets, particularly excelling in handling non-uniform hazy images.