Scholarlink

<p>In digital image processing, blur pixel identification and image enhancement are essential operations that strive to increase image sharpness and clarity. Traditional methods, such as edge detection and Wiener filtering, sometimes fail to handle complex blurs and can be computationally taxing. On the other hand, Deep Learning (DL) models provide improvements by learning from big datasets to better manage different types and conditions of blur. In this study, a novel DL-based method is proposed for image augmentation and blur pixel detection using a Modified Generative Adversarial Network (MGAN) and an Improved Deep Residual Network (IDRN). This method comprises three stages, including blur pixel identification using IDRN, deblurring of identified pixels with a statistical model, and final image enhancement and restoration using MGAN. Here, the results demonstrate that the MGAN model achieved state-of-the-art performance with an FOM of 0.919, PSNR of 38.276, SSIM of 0.931, and UQI of 0.935. This study underscores the potential of DL in advancing image restoration techniques, with applications in fields such as medical imaging, remote sensing, and surveillance.</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

Enhanced Image Restoration via Improved Deep Residual Network and Modified Generative Adversarial Network for Blur Pixel Identification

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • B. Baron Sam,
  • S. Vinson Joshua,
  • J. Jeba Christy

摘要

In digital image processing, blur pixel identification and image enhancement are essential operations that strive to increase image sharpness and clarity. Traditional methods, such as edge detection and Wiener filtering, sometimes fail to handle complex blurs and can be computationally taxing. On the other hand, Deep Learning (DL) models provide improvements by learning from big datasets to better manage different types and conditions of blur. In this study, a novel DL-based method is proposed for image augmentation and blur pixel detection using a Modified Generative Adversarial Network (MGAN) and an Improved Deep Residual Network (IDRN). This method comprises three stages, including blur pixel identification using IDRN, deblurring of identified pixels with a statistical model, and final image enhancement and restoration using MGAN. Here, the results demonstrate that the MGAN model achieved state-of-the-art performance with an FOM of 0.919, PSNR of 38.276, SSIM of 0.931, and UQI of 0.935. This study underscores the potential of DL in advancing image restoration techniques, with applications in fields such as medical imaging, remote sensing, and surveillance.