Scholarlink

<p>Images taken in low-light conditions experience significant degradation caused by inadequate illumination, resulting in diminished performance of both industrial and consumer devices. Enhancing low-light images is directed at enhancing the visual quality of pictures taken in environments with insufficient lighting. Nonetheless, these images frequently exhibit low visibility and noise, presenting a challenging task. Recently, notable advancements have been achieved through the adoption of deep learning methodologies. Current Retinex-based deep learning techniques are not yet optimal, as they fail to capitalize on valuable awareness from traditional methods. Additionally, the modification process is often either overly simplistic or excessively complex, leading to unsatisfactory real-world performance. To tackle these challenges, we introduce a new deep learning framework for low-light image enhancement. This investigation introduces an incorporated learning method multi-stage residual network (MSRNet) designed for enhancing low-light images. Multi-path residual network designs include a series of residual concatenation blocks stacked with Adaptive Residual Blocks. This structure captures spatial context information and extracts relevant features, enhancing the use of multi-level representations before the up-sampling phase. It also enables effective information and gradient flow within the network. The suggested framework additionally incorporates a novel attention mechanism called the Two fold Attention Module, aimed at optimizing the model’s representation capabilities. Our approach evaluated using two standard low-light datasets and benchmarked against seven state-of-the-art methods, achieving superior results in key quantitative metrics such as Peak Signal-to-Noise Ratio, Structural Similarity Index, Lightness Order Error, Natural Image Quality Evaluator, and Feature Similarity Index. The results highlight MSRNet’s ability to outperform existing methods across diverse conditions. The source code is available on GitHub at <a href="https://github.com/SATHISHMOTHE/MSRNet">https://github.com/SATHISHMOTHE/MSRNet</a>.</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

Multi-stage residual network with two fold attention mechanisms for low-light image enhancement

  • 期刊论文

  • 发表日期：

    2025年2月19日

查看全文
  • Sathish Mothe,
  • Srinivas Kankanala

摘要

Images taken in low-light conditions experience significant degradation caused by inadequate illumination, resulting in diminished performance of both industrial and consumer devices. Enhancing low-light images is directed at enhancing the visual quality of pictures taken in environments with insufficient lighting. Nonetheless, these images frequently exhibit low visibility and noise, presenting a challenging task. Recently, notable advancements have been achieved through the adoption of deep learning methodologies. Current Retinex-based deep learning techniques are not yet optimal, as they fail to capitalize on valuable awareness from traditional methods. Additionally, the modification process is often either overly simplistic or excessively complex, leading to unsatisfactory real-world performance. To tackle these challenges, we introduce a new deep learning framework for low-light image enhancement. This investigation introduces an incorporated learning method multi-stage residual network (MSRNet) designed for enhancing low-light images. Multi-path residual network designs include a series of residual concatenation blocks stacked with Adaptive Residual Blocks. This structure captures spatial context information and extracts relevant features, enhancing the use of multi-level representations before the up-sampling phase. It also enables effective information and gradient flow within the network. The suggested framework additionally incorporates a novel attention mechanism called the Two fold Attention Module, aimed at optimizing the model’s representation capabilities. Our approach evaluated using two standard low-light datasets and benchmarked against seven state-of-the-art methods, achieving superior results in key quantitative metrics such as Peak Signal-to-Noise Ratio, Structural Similarity Index, Lightness Order Error, Natural Image Quality Evaluator, and Feature Similarity Index. The results highlight MSRNet’s ability to outperform existing methods across diverse conditions. The source code is available on GitHub at https://github.com/SATHISHMOTHE/MSRNet.