Scholarlink

<p>Image enhancement and restoration are essential techniques aimed at improving the visual quality of images by correcting or modifying their attributes. Traditional methods, such as histogram equalization and Wiener filtering, often struggle with complex noise types and significant blurring, leading to over-smoothing and the loss of critical details. To address these limitations, this research proposes a novel methodology based on the Modified miRNet (MmiRNet) model for robust image enhancement and restoration. The approach is composed of three stages: preprocessing using Gaussian filtering, image inpainting with an Improved Contextual-Aware Deep Learning-based Convolutional Neural Network (CNN), and final enhancement through the MmiRNet model. Experimental validation on benchmark datasets, including MIAS Mammography (dataset 1), VHR-10_dataset_coco (dataset 2) and GPR1200 Dataset (dataset 3), demonstrates that MmiRNet significantly outperforms existing methods in key metrics such as accuracy, recall, F1-score, and Peak Signal-to-Noise Ratio (PSNR). With 90% of the training data, MmiRNet achieved a peak Figure of Merit (FOM) of 0.975, which is statistically significant compared to existing approaches, including MIRNet (0.854) and SqueezeNet (0.898). This research showcases the effectiveness of MmiRNet in providing advanced restoration capabilities, such as better handling of noise, fine-grained detail recovery, and structural integrity in heavily degraded 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

Enhanced image restoration via MmiRNet with context-aware deep learning inpainting

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • B. Baron Sam,
  • T. V. Geetha,
  • S. L. Soniya

摘要

Image enhancement and restoration are essential techniques aimed at improving the visual quality of images by correcting or modifying their attributes. Traditional methods, such as histogram equalization and Wiener filtering, often struggle with complex noise types and significant blurring, leading to over-smoothing and the loss of critical details. To address these limitations, this research proposes a novel methodology based on the Modified miRNet (MmiRNet) model for robust image enhancement and restoration. The approach is composed of three stages: preprocessing using Gaussian filtering, image inpainting with an Improved Contextual-Aware Deep Learning-based Convolutional Neural Network (CNN), and final enhancement through the MmiRNet model. Experimental validation on benchmark datasets, including MIAS Mammography (dataset 1), VHR-10_dataset_coco (dataset 2) and GPR1200 Dataset (dataset 3), demonstrates that MmiRNet significantly outperforms existing methods in key metrics such as accuracy, recall, F1-score, and Peak Signal-to-Noise Ratio (PSNR). With 90% of the training data, MmiRNet achieved a peak Figure of Merit (FOM) of 0.975, which is statistically significant compared to existing approaches, including MIRNet (0.854) and SqueezeNet (0.898). This research showcases the effectiveness of MmiRNet in providing advanced restoration capabilities, such as better handling of noise, fine-grained detail recovery, and structural integrity in heavily degraded images.