计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (19): 204-215.DOI: 10.3778/j.issn.1002-8331.1801-0472

• 图形图像处理 • 上一篇    下一篇

结合天空区域识别的单幅图像去雾方法

李尧羿1,杜宇超2,顾振飞3,4   

  1. 1.南京邮电大学 贝尔英才学院,南京 210003
    2.西交利物浦大学 计算机科学与软件工程系,江苏 苏州 215123
    3.南京邮电大学 物联网学院,南京 210003
    4.南京信息职业技术学院 电子信息学院,南京 210023
  • 出版日期:2018-10-01 发布日期:2018-10-19

Single image dehazing method via sky region recognition

LI Yaoyi1, DU Yuchao2, GU Zhenfei3,4   

  1. 1.Bell Honors School, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
    2.Department of Computer Science and Software Engineering, Xi’an Jiaotong-Liverpool University, Suzhou, Jiangsu 215123, China
    3.School of Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
    4.School of Electronic Information Engineering, Nanjing College of Information Technology, Nanjing 210023, China
  • Online:2018-10-01 Published:2018-10-19

摘要: 针对现有图像去雾方法易于在天空区域引入负面视觉效果的缺陷,提出一个结合天空区域识别的单幅图像去雾方法;提出一个新的天空区域特征先验知识,并利用所提先验将雾天降质图像分割为天空与非天空区域;基于天空区域对大气光进行估计,并利用暗通道先验和导向全变分模型对非天空区域的透射率进行估计,从而基于大气散射模型获得去雾处理后的图像;使用一种邻域自适应的Retinex方法克服了去雾处理后图像偏暗的问题。对比实验证明,所提方法相比现有的类似方法具备更好的有效性及鲁棒性。

关键词: 图像去雾, 天空区域特征先验, 暗通道先验, Retinex模型

Abstract: Considering that the sky region in a hazy image suffers from negative visual effect, a single image dehazing method via sky region recognition is proposed. Firstly, a simple but effective image prior-sky region feature is present, which is used to segment a hazy image into sky region and non-sky region. Then based on the recognized sky region, the atmospheric is obtained. By using the dark channel prior and further refined via the guided total variation model, the transmission of the non-sky region is estimated. Consequently, the clear image can be restored based on the atmospheric scattering model. Furthermore, a neighborhood adaptive Retinex method is also proposed to overcome the problem that image looks dim after haze removal. The comparison experimental results verify that the proposed method can produce results comparative to and even better than existing similar techniques with respect of the effectiveness and robustness.

Key words: image dehazing, sky region feature prior, dark channel prior, Retinex model