Scholarlink

<p>Image deblurring remains a challenging task in the field of computer vision, particularly in complex scenarios characterized by motion blur and noise. In this paper, we introduce ICTD, a novel deblurring network that integrates convolutional neural networks (CNN) and transformer architectures within a diffusion model framework. The specific content can be divided into two parts: firstly, our approach employs a multi-scale CNN as an initial deblurring module, followed by a transformer featuring an Inter–Intra attention mechanism that serves as the reverse denoising module of the diffusion model. This design enhances the initial deblurring results by incorporating finer details. Secondly, to further bolster robustness—especially against noise—we integrate an auxiliary convolutional module known as DeNet. Extensive experiments conducted on various blur datasets demonstrate that ICTD outperforms both classical and traditional deblurring methods, effectively addressing dynamic deblurring challenges in noisy environments. The integration of CNNs, Transformers, and diffusion models within ICTD highlights the adaptability and complementarity of these architectures, resulting in superior image restoration outcomes. The source code is available for research purposes at <a href="https://github.com/fengbuxingbujue/ICTD.git">https://github.com/fengbuxingbujue/ICTD.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

ICTD: integrating CNN and transformer with diffusion models for robust image deblurring and denoising

  • 期刊论文

  • 发表日期：

    2025年7月23日

查看全文
  • Haoyu Liu,
  • Guangmang Cui,
  • Jufeng Zhao,
  • Ziyi Chen,
  • Yucheng Huang,
  • Yuesheng Hao,
  • Junjie Ouyang

摘要

Image deblurring remains a challenging task in the field of computer vision, particularly in complex scenarios characterized by motion blur and noise. In this paper, we introduce ICTD, a novel deblurring network that integrates convolutional neural networks (CNN) and transformer architectures within a diffusion model framework. The specific content can be divided into two parts: firstly, our approach employs a multi-scale CNN as an initial deblurring module, followed by a transformer featuring an Inter–Intra attention mechanism that serves as the reverse denoising module of the diffusion model. This design enhances the initial deblurring results by incorporating finer details. Secondly, to further bolster robustness—especially against noise—we integrate an auxiliary convolutional module known as DeNet. Extensive experiments conducted on various blur datasets demonstrate that ICTD outperforms both classical and traditional deblurring methods, effectively addressing dynamic deblurring challenges in noisy environments. The integration of CNNs, Transformers, and diffusion models within ICTD highlights the adaptability and complementarity of these architectures, resulting in superior image restoration outcomes. The source code is available for research purposes at https://github.com/fengbuxingbujue/ICTD.git.