Scholarlink

<p>Image denoising restores the essential details of images corrupted by noise in the field of low-level vision and image processing. Especially, Convolutional Neural Networks (CNNs) provide excellent denoising performance on eliminating the spatially invariant noise. Despite the deeper network possessing high learning ability, some negative impacts often emerge, degrading the overall performance. As the network size gets deeper, the number of parameters and denoising time get augmented correspondingly, adding complexity to achieve the effective denoising performance. Hence, this research proposes the Upgraded Illumination Map enabled Dilated U-shaped Denoising Convolution Network with Predator–Prey-Hunter Optimization (UIM-DUDnCNN-PPHO) for effective noise removal. By leveraging UIM, the model adapts dynamically to diverse lighting conditions, ensuring effective noise reduction without compromising the image details. Specifically, the application of a Dilated U-shaped denoising convolutional neural network enhances the feature representation and flexibility, facilitating the model to learn the complementary image features, leading to a high-quality denoised image. Further, the incorporation of the PPHO algorithm enables the effective training of the proposed model via the selection of optimal hyperparameters, resulting in a reduction of the overall computation complexity. Experimental results show that the proposed UIM-DUDnCNN model achieves a PSNR of 50.96&#xa0;dB and an SSIM of 0.93 on the FMD dataset. Furthermore, across varying noise intensities, the model maintains high performance with a PSNR of 60.89&#xa0;dB and SSIM of 0.94, demonstrating its superiority in noise suppression compared to state-of-the-art approaches.</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

UIM-DUDnCNN-PPHO: Optimization Driven Upgraded Illumination Map Enabled Dilated U-Shaped Denoising Convolutional Neural Network for Image Denoising

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Jitendra Chandrakant Musale,
  • Amit A. Kadam

摘要

Image denoising restores the essential details of images corrupted by noise in the field of low-level vision and image processing. Especially, Convolutional Neural Networks (CNNs) provide excellent denoising performance on eliminating the spatially invariant noise. Despite the deeper network possessing high learning ability, some negative impacts often emerge, degrading the overall performance. As the network size gets deeper, the number of parameters and denoising time get augmented correspondingly, adding complexity to achieve the effective denoising performance. Hence, this research proposes the Upgraded Illumination Map enabled Dilated U-shaped Denoising Convolution Network with Predator–Prey-Hunter Optimization (UIM-DUDnCNN-PPHO) for effective noise removal. By leveraging UIM, the model adapts dynamically to diverse lighting conditions, ensuring effective noise reduction without compromising the image details. Specifically, the application of a Dilated U-shaped denoising convolutional neural network enhances the feature representation and flexibility, facilitating the model to learn the complementary image features, leading to a high-quality denoised image. Further, the incorporation of the PPHO algorithm enables the effective training of the proposed model via the selection of optimal hyperparameters, resulting in a reduction of the overall computation complexity. Experimental results show that the proposed UIM-DUDnCNN model achieves a PSNR of 50.96 dB and an SSIM of 0.93 on the FMD dataset. Furthermore, across varying noise intensities, the model maintains high performance with a PSNR of 60.89 dB and SSIM of 0.94, demonstrating its superiority in noise suppression compared to state-of-the-art approaches.