基于天空区域分割与置信度图导向融合的去雾方法研究
DOI:
CSTR:
作者:
作者单位:

(重庆邮电大学 数据工程与可视计算重点实验室,重庆 400065)

作者简介:

孙开伟 (1987-),男,博士,副教授,硕士生导师,主要从事机器学习、数据挖掘以及大数据分析方面的研究.

通讯作者:

中图分类号:

基金项目:

国家自然科学基金(61806033)、 重庆市教委项目(KJCXZD2020027)和重庆市自然科学基金面上项目(cstc2019jcyj-msxmX0021)资助项目


Dehazing algorithm based on sky region segmentation and reliability map guided fusion
Author:
Affiliation:

(Key Laboratory of Data Engineering and Visual Computing,Chongqing University of Posts and Telecommunications,Chongqing 400065, China)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    基于暗通道先验的去雾算法总是存在复原结果中天空区域处理不佳等问题,为了进一步优化对传输函数的估计,本文提出一种基于置信度图导向融合的传输函数优化方法。首先,将雾天图像的天空区域分离出来,以达到对天空区域的优化;计算窗口级暗通道与像素级暗通道,以平滑传输函数在物体边缘并保留小于窗口尺寸的细节特征;最后,计算窗口级暗通道与像素级暗通道之间的置信度图,以其为导向对两者进行融合得到优化的传输函数图,实现图像去雾。实验结果表明,本文算法可达到很好的复原结果优化效果。

    Abstract:

    The sky region in the restoration results for dehazing algorithm based on dark channel prior always exists the drawbacks such as halos. In order to further optimize the estimation of transmission function,this paper proposes a transmission function optimization method based on reliability map guided fusion.Firstly,the sky region of the hazy image is segmented and optimized;the window level dark level channel and pixel dark level channel are calculated to ensure the smoothness of the transmission function at the edge of the objects and the outstanding of the detail features smaller than the window size;finally,the reliability map between window level dark channel and pixel level dark channel is calculated.The refine transmission function map is obtained to realize image dehazing by the fusion of these two dark channels through the guidance of the reliability map.The experimental results show that the estimation of transmission function optimized by the proposed method could achieve fine image dehazing effects.

    参考文献
    相似文献
    引证文献
引用本文

孙开伟,冉雪,李彦,宣立德.基于天空区域分割与置信度图导向融合的去雾方法研究[J].光电子激光,2023,34(2):147~155

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2022-03-31
  • 最后修改日期:2022-05-25
  • 录用日期:
  • 在线发布日期: 2023-02-17
  • 出版日期:
文章二维码