Scholarlink

<p>Image dehazing is an active topic in computer vision. With the rapid development of deep learning, many image dehazing networks have been proposed. However, existing methods either lack lightweight attention mechanisms or are constrained by the limited ability of Transformers to effectively capture local spatial information, which can make it difficult to fully exploit available information and lead to the loss of important details. To overcome these limitations, this paper proposes a two-stage Generative Adversarial Dehazing Network Based on Attention Mechanism. The framework adopts a coarse-to-fine strategy: a symmetric encoder-decoder architecture integrated with Feature-Enhanced Residual Blocks (FERB) to effectively extract key features, while Dense Collaborative Attention Blocks (DCAB) enhance deep-level representation through multi-dimensional weighting. Subsequently, the refinement stage employs Multi-Scale Dilated Convolution Blocks (MDCB) and innovative Attention Feature Fusion Blocks (AFFB) to recover textures and edges and simultaneously enhance feature flow. Finally, a Multi-Scale Discriminator (MSD) is applied to supervise the restoration of the dehazed images. Experimental comparisons show that our method achieves comparable or superior results to existing state-of-the-art methods, effectively preserving image details while mitigating information loss on both synthetic and real-world datasets.</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

Two-stage generative adversarial dehazing network based on attention mechanism

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Yan Wang,
  • Shurong Li,
  • Huan Wang

摘要

Image dehazing is an active topic in computer vision. With the rapid development of deep learning, many image dehazing networks have been proposed. However, existing methods either lack lightweight attention mechanisms or are constrained by the limited ability of Transformers to effectively capture local spatial information, which can make it difficult to fully exploit available information and lead to the loss of important details. To overcome these limitations, this paper proposes a two-stage Generative Adversarial Dehazing Network Based on Attention Mechanism. The framework adopts a coarse-to-fine strategy: a symmetric encoder-decoder architecture integrated with Feature-Enhanced Residual Blocks (FERB) to effectively extract key features, while Dense Collaborative Attention Blocks (DCAB) enhance deep-level representation through multi-dimensional weighting. Subsequently, the refinement stage employs Multi-Scale Dilated Convolution Blocks (MDCB) and innovative Attention Feature Fusion Blocks (AFFB) to recover textures and edges and simultaneously enhance feature flow. Finally, a Multi-Scale Discriminator (MSD) is applied to supervise the restoration of the dehazed images. Experimental comparisons show that our method achieves comparable or superior results to existing state-of-the-art methods, effectively preserving image details while mitigating information loss on both synthetic and real-world datasets.