Scholarlink

<p>Image denoising based on deep learning has garnered widespread attention. However, the major of existing researches focus on denoising images in well-lit scenes, with very few studies addressing denoising in low-illumination environments. Therefore, this paper proposes a object attention inverted Transformer network (OA-iTNet) to tackle the issue of denoising low-light-level (LLL) images, which is composed of multi-scale feature fusion block, object detail enhancement attention module, and a serial combination of region feature block embedded with inverted Transformer residual block. To address the scarcity of LLL denoising datasets, this paper also introduces a real LLL image dataset, 3LI-D, captured by Multi-Pixel Photon Counter (MPPC) at <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="530_2025_1859_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="33" /> </InlineMediaObject> <EquationSource Format="TEX">\(10^{-3}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mn>10</mn> <mrow> <mo>-</mo> <mn>3</mn> </mrow> </msup> </math></EquationSource> </InlineEquation> lux. Experimental results demonstrate that this method not only outperforms existing deep convolutional neural network denoisers on LLL image dataset but also exhibits superior generalization performance on traditional denoising image datasets.</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

OA-iTNet: object attention inverted transformer network for low-light-level image denoising

  • 期刊论文

  • 发表日期：

    2025年6月1日

查看全文
  • Feng Wang,
  • Liju Yin,
  • Yiming Qin,
  • Xiaoning Gao,
  • Hui Zhou,
  • Yulin Deng

摘要

Image denoising based on deep learning has garnered widespread attention. However, the major of existing researches focus on denoising images in well-lit scenes, with very few studies addressing denoising in low-illumination environments. Therefore, this paper proposes a object attention inverted Transformer network (OA-iTNet) to tackle the issue of denoising low-light-level (LLL) images, which is composed of multi-scale feature fusion block, object detail enhancement attention module, and a serial combination of region feature block embedded with inverted Transformer residual block. To address the scarcity of LLL denoising datasets, this paper also introduces a real LLL image dataset, 3LI-D, captured by Multi-Pixel Photon Counter (MPPC) at \(10^{-3}\) 10 - 3 lux. Experimental results demonstrate that this method not only outperforms existing deep convolutional neural network denoisers on LLL image dataset but also exhibits superior generalization performance on traditional denoising image datasets.