Scholarlink

<p>IHRFNet (Integrating Residual and Hybrid Attention Blocks with Feature Extraction Network for Image Denoising) is introduced as a novel framework specifically designed for color image denoising. The model effectively suppresses noise while preserving fine textures and vivid color details by adopting a hybrid architecture that combines serial and parallel pathways with spatial and channel-wise attention mechanisms. Three specialized modules, Enhanced Residual Attention Block (ERAB), Hybrid Dilated Residual Attention Block (HDRAB), and Hybrid Sparse Attention Block (HSAB), jointly improve local detail modeling, expand receptive fields, and improve efficiency. Extensive evaluations of benchmark datasets demonstrate the superiority of the proposed framework. In synthetic Gaussian noise datasets (CBSD68, Kodak24 and McMaster), IHRFNet achieves 1.8 to 3.2% higher PSNR and 1.5 to 2.7% higher SSIM than state-of-the-art methods at different noise levels. In real-world benchmarks, the framework records 39.46 dB / 0.916 SSIM on SIDD and 39.58 dB / 0.953 SSIM on DND, placing it among the top-performing models. These results confirm the model’s ability to restore natural textures, structural details, and color fidelity under both synthetic and real noise conditions. Although the integration of multiple attention modules slightly increases computational complexity, the significant improvements in image quality justify this trade-off. In general, IHRFNet emerges as an efficient and robust solution for practical image denoising, with promising potential for extension to tasks such as color superresolution and joint denoising-enhancement. The implementation of IHRFNet is publicly accessible at <a href="https://github.com/akhilarajd/IHRFNet">https://github.com/akhilarajd/IHRFNet</a>, offering a valuable resource for future research and practical deployment.</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

IHRFNet: Integrating Residual and Hybrid Attention Blocks with Feature Extraction Network for Image Denoising

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Akhilaraj D,
  • Joseph Zacharias

摘要

IHRFNet (Integrating Residual and Hybrid Attention Blocks with Feature Extraction Network for Image Denoising) is introduced as a novel framework specifically designed for color image denoising. The model effectively suppresses noise while preserving fine textures and vivid color details by adopting a hybrid architecture that combines serial and parallel pathways with spatial and channel-wise attention mechanisms. Three specialized modules, Enhanced Residual Attention Block (ERAB), Hybrid Dilated Residual Attention Block (HDRAB), and Hybrid Sparse Attention Block (HSAB), jointly improve local detail modeling, expand receptive fields, and improve efficiency. Extensive evaluations of benchmark datasets demonstrate the superiority of the proposed framework. In synthetic Gaussian noise datasets (CBSD68, Kodak24 and McMaster), IHRFNet achieves 1.8 to 3.2% higher PSNR and 1.5 to 2.7% higher SSIM than state-of-the-art methods at different noise levels. In real-world benchmarks, the framework records 39.46 dB / 0.916 SSIM on SIDD and 39.58 dB / 0.953 SSIM on DND, placing it among the top-performing models. These results confirm the model’s ability to restore natural textures, structural details, and color fidelity under both synthetic and real noise conditions. Although the integration of multiple attention modules slightly increases computational complexity, the significant improvements in image quality justify this trade-off. In general, IHRFNet emerges as an efficient and robust solution for practical image denoising, with promising potential for extension to tasks such as color superresolution and joint denoising-enhancement. The implementation of IHRFNet is publicly accessible at https://github.com/akhilarajd/IHRFNet, offering a valuable resource for future research and practical deployment.