Scholarlink

<p>Neutron radiographic images (NRIs) typically suffer from multiple distortions, including various types of noise, geometric unsharpness, and white spots. Image quality assessment (IQA) can guide on-site image screening and even provide metrics for subsequent image processing. However, existing IQA methods for NRIs cannot effectively evaluate the quality of real NRIs with a specific distortion of white spots, limiting their practical application. In this paper, a novel no-reference IQA method is proposed to comprehensively evaluate the quality of real NRIs with multiple distortions. First, we construct large-scale NRI datasets with more than 20,000 images, including high-quality original NRIs and synthetic NRIs with various distortions. Next, an image quality calibration method based on visual salience and a local quality map is introduced to label the NRI dataset with quality scores. Finally, a lightweight convolutional neural network (CNN) model is designed to learn the abstract relationship between the NRIs and quality scores using the constructed NRI training dataset. Extensive experimental results demonstrate that the proposed method exhibits good consistency with human visual perception when evaluating both real NRIs and processed NRIs using enhancement and restoration algorithms, highlighting its application potential.</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

Comprehensive quality assessment method for neutron radiographic images based on CNN and visual salience

  • 期刊论文

  • 发表日期：

    2025年5月8日

查看全文
  • Zhen Zhang,
  • Cheng-Bo Meng,
  • Xiao-Long Jiang,
  • Chen-Yi Zhao,
  • Shuang Qiao,
  • Tian Zhang

摘要

Neutron radiographic images (NRIs) typically suffer from multiple distortions, including various types of noise, geometric unsharpness, and white spots. Image quality assessment (IQA) can guide on-site image screening and even provide metrics for subsequent image processing. However, existing IQA methods for NRIs cannot effectively evaluate the quality of real NRIs with a specific distortion of white spots, limiting their practical application. In this paper, a novel no-reference IQA method is proposed to comprehensively evaluate the quality of real NRIs with multiple distortions. First, we construct large-scale NRI datasets with more than 20,000 images, including high-quality original NRIs and synthetic NRIs with various distortions. Next, an image quality calibration method based on visual salience and a local quality map is introduced to label the NRI dataset with quality scores. Finally, a lightweight convolutional neural network (CNN) model is designed to learn the abstract relationship between the NRIs and quality scores using the constructed NRI training dataset. Extensive experimental results demonstrate that the proposed method exhibits good consistency with human visual perception when evaluating both real NRIs and processed NRIs using enhancement and restoration algorithms, highlighting its application potential.