系统工程与电子技术 ›› 2023, Vol. 45 ›› Issue (6): 1606-1615.doi: 10.12305/j.issn.1001-506X.2023.06.03

• 电子技术 • 上一篇    

基于图像天空区域分割的去雾算法

孙景荣1,2, 陈哲哲1, 谢林昌1, 杜梦欣1, 宋诗斌3,*   

  1. 1. 西安电子科技大学空间科学与技术学院, 陕西 西安 710071
    2. 近地面探测技术重点实验室, 江苏 无锡 214000
    3. 山东科技大学机器人与智能技术重点实验室, 山东 青岛 266590
  • 收稿日期:2022-01-24 出版日期:2023-05-25 发布日期:2023-06-01
  • 通讯作者: 宋诗斌
  • 作者简介:孙景荣(1975—), 女, 副教授, 硕士研究生导师, 博士, 主要研究方向为低照度图像处理与分析、信号检测与信息处理
    陈哲哲(1997—), 男, 硕士研究生, 主要研究方向为图像处理、计算机视觉
    谢林昌(1994—), 男, 硕士研究生, 主要研究方向为计算机视觉
    杜梦欣(1998—), 女, 硕士研究生, 主要研究方向为智能交通系统研究
    宋诗斌(1988—), 男, 讲师, 博士, 主要研究方向为智能机器人技术、机器人定位导航

Haze removal algorithm based on image sky region segmentation

Jingrong SUN1,2, Zhezhe CHEN1, Linchang XIE1, Mengxin DU1, Shibin SONG3,*   

  1. 1. School of Aerospace Science and Technology, Xidian University, Xi'an 710071, China
    2. Science and Technology on Near-Surface Defection Laboratory, Wuxi 214000, China
    3. Shandong Provincial Key Laboratory of Robot and Intelligent Technology, Shandong University of Science and Technology, Qingdao 266590, China
  • Received:2022-01-24 Online:2023-05-25 Published:2023-06-01
  • Contact: Shibin SONG

摘要:

为了提升图像去雾效果, 提出了基于天空区域分割的透射率合成图像去雾算法, 通过雾霾图像构造的概率分布函数, 结合迭代阈值分割及天空区域的暗、亮通道以及最大连通性确定出粗分割阈值。利用引导滤波加强天空与非天空区域像素灰度的差异性,实现了精准分割。利用对数自适应变换估计了天空区域透射率, 利用改进的暗通道先验算法估计了非天空区域透射率, 通过合成对应像素的透射率实现了图像的去雾。实验结果表明, 相比其他去雾算法, 所提算法在各客观指标上均有所提高, 对天空与非天空区域分割准确, 去雾图像整体视觉效果良好。

关键词: 天空分割, 图像去雾, 暗通道先验, 透射率映射

Abstract:

In order to improve the haze removal effect of images, an image haze removal algorithm based on transmissivity synthesis of sky region segmentation is proposed in this paper. The rough segmentation threshold is determined based on the probability distribution function constructed by haze image with the combination of iterative threshold segmentation with dark and bright channels and the maximum connectivity. Guided filtering is used to enhance the difference of the pixel gray between sky and non-sky regions to achieve the accurate segmentation. Logarithmic adaptive transformation is adopted to estimate the transmissivity in sky region, and the improved dark channel prior algorithm is used to estimate the transmissivity in non-sky region. The image was defogged by synthesizing the transmissivity of the corresponding pixels. Experimental results show that compared with other defogging algorithms, the algorithm presented in this paper achieves improvements in all objective indexes, which accurately segments sky and non-sky regions, and achieves good overall visual effect of defogging images.

Key words: sky segmentation, image defogging, dark channel prior, transmittance mapping

中图分类号: