Scholarlink

<p>This research introduces a LVSIR-Pix2Pix framework whose objective is to restore images captured under non-ideal conditions. This approach mitigates background noise and reconstructs lost laser contour information, thereby enhancing the accuracy of obtaining three-dimensional information in complex scenarios. Compared to Gaussian-based and Meanshift-based methods, the proposed approach exhibits a mean squared error (MSE) reduction of 0.243 and 0.223, with corresponding percentage decreases of 72.66% and 66.97%, respectively, in the detail space 𝔻. Additionally, we introduce a loss function that combines global and detail spatial components. When used simultaneously, this loss function achieves an average MSE reduction of 0.125, corresponding to a 68.1% decrease compared to the original MSE.</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

A novel image restoration method for laser vision sensors to support automatic grinding and polishing

  • 期刊论文

  • 发表日期：

    2025年12月17日

查看全文
  • Shibo Liu,
  • Jing Zhang,
  • Xiaoqi Wang,
  • Christophe Claramunt,
  • Gang Tang,
  • Yanling Xu,
  • Huajun Zhang

摘要

This research introduces a LVSIR-Pix2Pix framework whose objective is to restore images captured under non-ideal conditions. This approach mitigates background noise and reconstructs lost laser contour information, thereby enhancing the accuracy of obtaining three-dimensional information in complex scenarios. Compared to Gaussian-based and Meanshift-based methods, the proposed approach exhibits a mean squared error (MSE) reduction of 0.243 and 0.223, with corresponding percentage decreases of 72.66% and 66.97%, respectively, in the detail space 𝔻. Additionally, we introduce a loss function that combines global and detail spatial components. When used simultaneously, this loss function achieves an average MSE reduction of 0.125, corresponding to a 68.1% decrease compared to the original MSE.